Multi-agent scientific research system

Abstract

Provided are computing systems and methods that can improve the efficiency and effectiveness of scientific research. Traditional methods can struggle with the rapid evaluation and generation of research hypotheses due to the vast amount of scientific literature and the complexity of integrating multifaceted scientific data. In contrast, example systems provided herein, which can be referred to as a multi-agent scientific research system, can utilize advanced machine learning models to automate and optimize these processes.

Description

MULTI-AGENT SCIENTIFIC RESEARCH SYSTEMPRIORITY CLAIM

[0001] The present application is based on and claims priority to United States Provisional Application Number 63 / 754,109 having a filing date of February 5, 2025 and United States Provisional Application Number 63 / 773,359 having a filing date of March 17, 2025. The present application claims priority to and the benefit of each of such applications and hereby incorporates all such applications herein by reference in their entirety.FIELD

[0002] The present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to a multi-agent scientific research system that leverages machine learning models to perform scientific research with improved computational efficiency.BACKGROUND

[0003] Generic computational systems can encounter technical challenges in managing extensive datasets typically associated with scientific research. These systems can struggle with the high computational demands required for efficient analysis of large datasets. This can include issues in real-time data processing, integrating diverse data types, and adapting to new scientific findings. As one example, in generating and evaluating scientific hypotheses, traditional systems can process extensive data volumes repetitively without efficiently prioritizing information that can yield important scientific insights, thereby increasing computational overhead and inefficiencies.

[0004] Consequently, there is a need to enhance the computational efficiency and effectiveness of systems used in scientific research to more strategically manage, process, and evaluate large and complex datasets. Specifically, a generic computational system not designed for scientific research can exhibit decreased efficiency in evaluating and ranking various hypotheses. Without specialized algorithms that can assess hypotheses based on scientific validity, relevance, and potential impact, the computing system might allocate considerable time and resources to less promising inquiries.SUMMARY

[0005] A system of one or more computers can be configured to perform particular operations or actions by virtue of having software, firmware, hardware, or a combination of them installed on the system that in operation causes or cause the system to perform the actions. One or more computer programs can be configured to perform particular operations or actions by virtue of including instructions that, when executed by data processing apparatus, cause the apparatus to perform the actions.

[0006] One general aspect includes a computing system for efficiently performing scientific research. The computing system also includes one or more processors; and one or more non-transitory computer-readable media that collectively store computer-readable instructions that, when executed by the one or more processors, cause the computing system to implement a multi-agent scientific research system, the multi -agent scientific research system may include: a hypothesis generation agent configured to process data descriptive of a scientific research goal with one or more machine-learned models to automatically generate a plurality of research hypothesis statements expressed in a natural language; and an evolution agent configured to iteratively process at least some of the plurality of research hypothesis statements with one or more machine-learned models to perform a tournament-based evaluation that generates a ranking of the plurality of research hypothesis statements. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0007] Implementations may include one or more of the following features. The computing system of any preceding claim, where the multi -agent scientific research system is configured to output a research proposal for the scientific research goal, where the research proposal may include one or more of the plurality of research hypothesis statements that have been selected based at least in part on the ranking of the plurality of research hypothesis statements. The multi-agent scientific research system is configured to: receive textual input descriptive of the scientific research goal; and process the textual input with one or more machine-learned models to generate a configuration plan that describes the scientific research goal and one or more of: constraints, considerations, information sources, or criteria for success; where the hypothesis statement generation agent is configured to process the configuration plan to generate the plurality of research hypothesis statements. The hypothesis statement generation agent is configured to perform zero-shot prompting to generate a plurality of source hypothesis statements. The hypothesis statement generation agent isconfigured to perform web search, information retrieval, or other tool use to obtain a plurality of scientific research articles and generate a plurality of source hypothesis statements based on the plurality of scientific research articles. The hypothesis statement generation agent is configured to perform a self-play technique to propose, debate, and / or refine a source hypothesis statement to generate one of the research hypothesis statements. The hypothesis statement generation agent is configured to perform a self-play technique in which the one or more machine-learned models of the hypothesis statement generation agent execute a multiturn natural language evaluation of a source hypothesis statement to generate one of the research hypothesis statements. The hypothesis statement generation agent is configured to generate one or more intermediate assumptions for inclusion in one or more of the plurality of research hypothesis statements. The multi-agent scientific research system further may include a reflection agent configured to: perform web search, information retrieval, or other tool use to retrieve one or more academic research articles; and perform, with one or more machine-learned models, model reasoning over the one or more of the plurality of research hypothesis statements in view of the one or more academic research articles. The model reasoning may include, for each research hypothesis statement: identifying one or more sets of experimental results or physical observations from the one or more academic research articles; and determining a likelihood that the research hypothesis statement may include a logical cause of the one or more sets of experimental results or physical observations. The multi-agent scientific research system further may include a reflection agent configured to perform a deep verification process on or more of the plurality of research hypothesis statements using one or more machine-learned models, where the deep verification process may include: identifying one or more assumptions or sub-assumptions contained in or implicit from the research hypothesis statement; and evaluating a correctness of each of the one or more assumptions or sub-assumptions. The evolution agent performs an iterative pairwise comparison of the plurality of research hypothesis statements. The evolution agent processes two or more of the plurality of research hypothesis statements and a fixed zero-shot prompt with the one or more machine-learned models to generate a comparison of two or more of the plurality of research hypothesis statements. The evolution agent controls the one or more machine-learned models to perform a multi-turn natural language debate about merits of two or more of the plurality of research hypothesis statements. The multi-agent scientific research system further may include an improve agent configured to process one or more of the plurality of research hypothesis statements with one or more machine-learned models to generate one or more modifications to the one or more of the plurality of researchhypothesis statements. The one or more modifications may include modifying one of the research hypothesis statements based on one or more other, top-ranking research hypothesis statements. The one or more modifications simplify one or more research hypothesis statements, expand one or more research hypothesis statements, add detail to one or more research hypothesis statements, improve the practicality of one or more research hypothesis statements, improve the grounding of one or more research hypothesis statements, or increase the novelty of one or more research hypothesis statements. The multi-agent scientific research system further may include a meta-review agent configured to process one or more of the plurality of research hypothesis statements and a plurality of academic research articles with one or more machine-learned models to identify one or more common failure points described in one or more of the academic research articles that are not addressed by the one or more research hypothesis statements. The multi-agent scientific research system further may include a proximity graph agent configured to build a proximity graph between at least a subset of plurality of research hypothesis statements. The multi-agent scientific research system further may include a supervisor agent configured to orchestrate worker devices to execute the hypothesis generation agent and the evolution agent. The supervisor agent is configured to set a weight for each agent in the multi-agent scientific research system and to sample agent actions from each agent in the multi-agent scientific research system in proportion to its respective weight. The hypothesis generation agent is configured to automatically generate the plurality of research hypothesis statements in parallel. The evolution agent is configured to perform the tournament-based evaluation over multiple pairwise evaluations executed in parallel. Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium.

[0008] One general aspect includes a computer-implemented method. The computer-implemented method includes receiving textual input descriptive of a scientific research goal; and processing the textual input with one or more machine-learned models to generate a configuration plan that describes the scientific research goal and one or more of constraints, considerations, information sources, or criteria for success; processing the configuration plan with one or more machine-learned models to generate a plurality of research hypothesis statements; and iteratively performing a pairwise ranking process, each iteration of the pairwise ranking process may include: sampling two of the research hypothesis statements; and processing the two research hypothesis statements with one or more machine-learned models to select a higher ranked statement and a lower ranked statement. Other embodimentsof this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0009] Implementations may include one or more of the following features. The computer-implemented method where processing the configuration plan with one or more machine-learned models to generate a plurality of research hypothesis statements may include, for each of a plurality of generative iterations: controlling the one or more machine-learned models to perform a multi-turn self-play process that iteratively refines one of the research hypothesis statements. For each iteration of the pairwise ranking process, processing the two research hypothesis statements with the one or more machine-learned models may include controlling the one or more machine-learned models to perform a multi-turn natural language debate about merits of the two research hypothesis statements. Iteratively performing the pairwise ranking process may include performing at least two iterations of the pairwise ranking process in parallel. Iteratively performing the pairwise ranking process may include: at a first iteration, storing one or more intermediate results of the pairwise ranking process in a cache; and at a second, subsequent iteration, retrieving the one or more intermediate results from the cache and performing continued evaluation from the one or more intermediate results. Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 illustrates a flowchart of an example research process according to example implementations of aspects of the present disclosure;

[0011] Figure 2 presents a flowchart of an example user interaction and hypothesis development workflow according to example implementations of aspects of the present disclosure;

[0012] Figure 3 illustrates a block diagram of an example multi-agent system architecture according to example implementations of aspects of the present disclosure;

[0013] Figure 4 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0014] Figure 5 is a block diagram of an example processing flow for using machine-learned model(s) to process input(s) to generate output(s) according to example implementations of aspects of the present disclosure;

[0015] Figure 6 is a block diagram of an example sequence processing model according to example implementations of aspects of the present disclosure;

[0016] Figure 7 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example implementations of aspects of the present disclosure;

[0017] Figure 8 is a block diagram of an example model development platform according to example implementations of aspects of the present disclosure;

[0018] Figure 9 is a block diagram of an example training workflow for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0019] Figure 10 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example implementations of aspects of the present disclosure;

[0020] Figure 11 is a block diagram of an example networked computing system according to example implementations of aspects of the present disclosure;

[0021] Figure 12 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure; and

[0022] Figure 13 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure.DETAILED DESCRIPTION

[0023] Example aspects of the present disclosure are directed to computing systems and methods that can improve the efficiency and effectiveness of scientific research.Traditional methods can struggle with the rapid evaluation and generation of research hypotheses due to the vast amount of scientific literature and the complexity of integrating multifaceted scientific data. In contrast, example systems provided herein, which can be referred to as a multi-agent scientific research system, can utilize advanced machine learning models to automate and optimize these processes.

[0024] Specifically, the proposed system can include a hypothesis generation agent configured to process data descriptive of a scientific research goal with one or more machine-learned models to automatically generate a plurality of research hypothesis statements expressed in a natural language. For example, the hypothesis generation agent within the system can process a user-defined scientific goal to produce multiple research hypotheses.

[0025] The proposed system can also include an evolution agent configured to iteratively process at least some of the plurality of research hypothesis statements with one or more machine-learned models to perform a tournament-based evaluation that generates a ranking of the plurality of research hypothesis statements. For example, the proposed hypothesis statements can then be refined and evaluated through a pairwise tournament-based approach managed by the evolution agent. Together, the hypothesis generation agent and the evolution agent can improve the speed of the hypothesis generation and ranking process.

[0026] Furthermore, in some implementations, the system can include a reflection agent that utilizes machine-learned models to review generated hypotheses against extensive databases and scientific literature. This can improve the likelihood that the hypotheses are innovative and grounded in current scientific understanding. In some implementations, the reflection agent can identify experimental results and observations relevant to the hypotheses, offering a mechanism that can enhance their validity and potential for empirical testing.

[0027] Additionally, in some implementations, the multi-agent system can include an improvement agent that performs iterative refinement of hypothesis statements by suggesting modifications based on feedback from the evaluation processes. In some implementations, this feature can support continuous improvement and adaptation of research hypotheses.

[0028] In some implementations, each agent can be powered by or otherwise leverage one or more machine-learned models. For example, the models can be sequence processing models (e.g., Transformer Models), as described further herein. For example, the models can be large models such as large language models (LLMs), large multi-modal models (LMMs), or similar. The agents can use the same model or can use different models. The models can be expert models or can be generalist models. Models can be controlled or conditioned using one or more “prompts”. Prompts can include natural language instructions that instruct a model to perform a particular task. Example prompts are extensively described in the attached Appendix.

[0029] Thus, the present disclosure provides a system that can assist human researchers in generating, evaluating, and refining research hypotheses. By utilizing advanced machine learning models in a purpose-built multi-agent architecture, the system can reduce the computational effort needed for scientific research, and can also improve the speed of discovery in various scientific fields.

[0030] More particularly, a computing system designed to improve scientific research efficiency can include one or more processors coupled with one or more non-transitory computer-readable media. These components can store instructions that, when executed,enable the system to deploy a multi-agent scientific research system. In the context of computing and artificial intelligence, an “agent” can refer to a computerized system that is designed to act autonomously or semi-autonomously on behalf of a user or another program. Agents are programmed to make decisions and perform actions based on the environment they operate in and the objectives they are given. Agents can be implemented using a software, firmware, hardware, or a combination thereof.

[0031] In some implementations, the multi-agent scientific research system can receive textual input (e.g., from a human user such as a human scientist) that describes a specific scientific research goal. This input can be processed using one or more machine-learned models to create a detailed configuration plan. This plan outlines the research goal and can include various elements such as potential constraints that might affect the research, considerations to be taken into account, sources of information that could be useful, and / or criteria that can define the success of the research endeavor. For instance, if a researcher inputs a goal related to cancer research, the system can generate a configuration plan that includes constraints like available laboratory equipment, considerations such as ethical approvals, information sources like existing cancer research databases, and / or success criteria like identifying potential drug targets.

[0032] In some implementations, the multi-agent research system can include a hypothesis generation agent that can process data descriptive of a scientific research goal using one or more machine-learned models to automatically generate multiple research hypotheses statements in natural language. For example, the hypothesis statement generation agent can use the configuration plan to generate a diverse set of research hypothesis statements. As one example, given a research goal related to understanding the genetic factors of a disease, the hypothesis generation agent can use machine learning models to analyze existing genetic data and relevant scientific literature. From this analysis, the agent can then produce several plausible hypotheses about potential genetic markers or interactions that might influence the disease. This capability can improve the efficiency with which researchers can obtain well-founded hypotheses that can guide further empirical research.

[0033] In some implementations, the hypothesis statement generation agent can use zero-shot prompting techniques to generate a plurality of source hypothesis statements. This method can include employing pre-trained machine learning models that do not require taskspecific training data for each new task. This capability can be beneficial in scenarios where rapid hypothesis generation is desired across diverse scientific fields, enabling researchers to explore a range of potential research directions without extensive preliminary data gathering.

[0034] In some implementations, the hypothesis statement generation agent can use web search, information retrieval, and / or other tools to access and analyze scientific research articles. This capability allows the hypothesis statement generation agent to derive various source hypothesis statements from the content of these articles. For example, when tasked with generating hypotheses related to Alzheimer’s disease, the agent can search and retrieve recent scholarly articles on neurological studies, genetic research, or pharmaceutical trials related to Alzheimer’s. From the insights gathered, the agent can then formulate multiple preliminary hypotheses about potential causes, progression markers, or therapeutic targets for the disease. This process not only enriches the hypothesis generation with a broad spectrum of scientific knowledge but also improves the likelihood that the hypotheses are grounded in the latest research findings.

[0035] In some implementations, the hypothesis statement generation agent can use a self-play technique to propose, debate, and / or refine a source hypothesis statement, ultimately generating one of the research hypothesis statements. This technique can include executing a dialogue or debate between multiple instances of machine-learned models. Each instance can play a role in challenging and defending various aspects of a hypothesis. For example, one model can propose a hypothesis based on a set of data, while another model critiques its validity or suggests modifications based on additional data or differing interpretations. This iterative self-play process can improve the hypothesis, making it more robust and well-rounded before it is finalized.

[0036] Thus, in some implementations, the hypothesis statement generation agent can be configured to utilize a self-play technique where one or more machine-learned models can execute a multi-turn natural language evaluation of a source hypothesis statement. This process can include one or more models engaging in a simulated dialogue or debate, where they iteratively question, critique, and refine the hypothesis based on logical reasoning and available data. Such a self-play mechanism can be beneficial for exploring different perspectives and improving the likelihood that the generated hypotheses are not only innovative but also evaluated for their scientific merit.

[0037] In some implementations, the hypothesis statement generation agent can be configured to generate one or more intermediate assumptions that can be included in the research hypothesis statements. This feature allows the system to construct layered and more nuanced hypotheses, which are beneficial for addressing complex scientific questions. For example, when exploring potential treatments for a disease, the hypothesis generation agent can generate an intermediate assumption about the interaction between a specific drug and adisease pathway. This assumption can then be used to form a detailed hypothesis about the drug’s efficacy, which can guide experimental design and further research. The capability to generate and incorporate intermediate assumptions can improve the depth and relevance of the hypotheses produced.

[0038] In some implementations, the multi-agent system can include a reflection agent. The reflect agent can perform web searches, information retrieval, and / or utilize other tools to fetch one or more academic research articles relevant to the generated research hypotheses. This reflection agent can be configured to use machine learning models to analyze the content of these articles. For instance, if a hypothesis is related to neurodegenerative diseases, the reflection agent can retrieve and process the latest research articles about genetic factors influencing such conditions. This process can improve the likelihood that the hypotheses are not only innovative but also grounded in the latest scientific research, thereby enhancing the reliability and relevance of the hypotheses generated by the system.

[0039] In some implementations, the reflection agent can perform model reasoning on each research hypothesis statement. This can include identifying sets of experimental results or physical observations from the retrieved academic articles and assessing the likelihood that a given hypothesis could logically explain these observations. For example, if a hypothesis suggests a new molecular pathway included in Alzheimer’s disease, the reflection agent can analyze related articles to determine if there is empirical evidence supporting this pathway’s involvement in the disease. By evaluating the connection between the hypothesis and existing scientific data, the system can provide researchers with insights into the potential validity of their hypotheses. This can improve the formulation of more targeted and effective research strategies.

[0040] Furthermore, in some implementations, the reflection agent can be configured to perform a deep verification process on one or more of the generated research hypothesis statements using machine-learned models. This process can include identifying any assumptions or sub-assumptions that are either explicitly stated or implied within the hypothesis statements. For example, if a hypothesis suggests a specific protein interaction in a biological process, the reflection agent can identify this interaction as an assumption.Subsequently, the system can evaluate the correctness of each identified assumption or subassumption. This evaluation can include checking the assumption against known scientific databases or literature to verify its validity or plausibility. Such a verification process canimprove the likelihood that the hypotheses are not only innovative but also scientifically plausible.

[0041] In some implementations, the multi-agent scientific research system can include an evolution agent configured to iteratively process at least some of the generated research hypothesis statements. This agent can use one or more machine-learned models to conduct a tournament-based evaluation, ranking the hypotheses according to their potential validity and / or relevance to the defined research goals. For example, the evolution agent can use criteria such as novelty, scientific rigor, and / or alignment with current research to evaluate each hypothesis. Through iterative comparisons and rankings, the system can improve the prioritization of hypotheses that are more likely to yield significant scientific insights, thereby optimizing the research direction before empirical validation steps are taken. This process can enhance the efficiency of research planning and can assist in focusing resources (e.g., computational resources, laboratory resources, etc.) on the most promising hypotheses.

[0042] In some implementations, the evolution agent can be configured to perform an iterative pairwise comparison of the plurality of research hypothesis statements. This method can include selecting pairs of hypotheses and systematically comparing them to determine which hypothesis is more robust or relevant based on predefined criteria. For example, in one iteration, the system can compare two hypotheses regarding the effectiveness of a new drug compound. The evolution agent evaluates factors such as the underlying scientific basis, potential impact, and alignment with existing research. Through this iterative process, the system can rank hypotheses. This pairwise comparison approach can assist in iteratively refining the pool of hypotheses, potentially improving the focus of subsequent research efforts towards inquiries that are scientifically sound and likely to be fruitful.

[0043] Specifically, as one example approach for performing pairwise evaluation, the evolution agent can process two or more of the plurality of research hypothesis statements alongside a fixed zero-shot prompt, utilizing one or more machine-learned models to generate a comparison of these statements. This process can include the application of a standardized prompt that does not change between sessions, which can improve consistent evaluation criteria across different sets of hypotheses. For example, the fixed zero-shot prompt can be used to assess the hypotheses based on their innovativeness and feasibility within a specific scientific field. By applying this uniform prompt, the system can impartially compare multiple hypotheses, aiding in the identification of the most promising hypotheses based on objective, predefined metrics.

[0044] As another example approach for performing pairwise evaluation, the evolution agent can control one or more machine-learned models to perform a multi-turn natural language debate about the merits of two or more of the plurality of research hypothesis statements. This approach can allow the system to simulate a detailed scholarly discussion, where different hypotheses are evaluated against each other through a series of argumentative exchanges. For example, if two hypotheses propose different mechanisms for a biological process, the system can generate arguments for and against each hypothesis based on existing scientific knowledge, data accuracy, and logical reasoning. This debate can help in identifying the strengths and weaknesses of each hypothesis, thereby providing deeper insight into which hypothesis might hold more scientific validity or practical application. This approach can enhance the robustness of the selection process and can mimic the peer review mechanism typically used in scientific research fields.

[0045] In some implementations, the multi-agent scientific research system can also include an improve agent that is configured to process and refine one or more of the previously generated research hypothesis statements. This agent can use one or more machine-learned models to analyze the content and structure of each hypothesis, identifying areas where modifications can enhance clarity, relevance, and / or scientific rigor. For instance, the improve agent can modify a hypothesis by incorporating additional scientific data that has become available, potentially increasing the hypothesis’s accuracy and depth. Alternatively, the agent might rephrase the hypothesis to make it more concise and understandable, which can improve its communicability to other researchers and stakeholders.

[0046] As another example, the improvement agent can review top-ranking hypotheses and integrate successful elements into other hypotheses that have potential but can be lacking in certain areas. For example, if a top-ranking hypothesis introduces a novel methodological approach that is beneficial, the improvement agent can adapt this approach for use in other hypotheses where similar methods might improve outcomes. As yet another example, the improvement agent can expand upon hypotheses that are overly narrow, adding detail where it is needed to fully capture the implications of the research or to meet the criteria for experimental validation. This iterative improvement process not only refines the hypotheses but also can improve their grounding in the latest research and their ability to contribute to the field of study.

[0047] In some implementations, the multi-agent scientific research system can also include a meta-review agent configured to process one or more of the plurality of researchhypothesis statements alongside a plurality of academic research articles using one or more machine-learned models. This agent can identify common failure points described in the academic research articles that are not addressed by the current hypothesis statements. For example, if the hypothesis statements relate to a new therapeutic approach for a disease, the meta-review agent can analyze existing literature to pinpoint where similar approaches have failed in the past, such as issues with drug delivery mechanisms or unexpected side effects. This capability allows the system to refine hypothesis statements by incorporating lessons learned from past research, which can improve the robustness and potential success of future experimental validations.

[0048] In some implementations, the multi-agent scientific research system can also include a proximity graph agent configured to construct a proximity graph between at least a subset of the plurality of research hypothesis statements. The proximity graph can illustrate the relationships and similarities between different hypotheses. For example, when the system generates multiple hypotheses concerning the molecular mechanisms of a specific disease, the proximity graph agent can illustrate the connections between these hypotheses based on shared keywords, underlying theories, and / or referenced studies. This feature can be beneficial for researchers to identify clusters of closely related hypotheses, which can suggest a more focused area for detailed investigation or experimentation.

[0049] In some implementations, the multi-agent scientific research system can also include a supervisor agent that coordinates the operation of worker devices to execute or effectuate the agents described herein. These worker devices can be configured to execute specific tasks assigned by or corresponding to each agent. For example, one worker device can be tasked with generating initial hypothesis statements based on a set research goal, while another can handle the evaluation and ranking of these hypotheses. This division of labor, coordinated by the supervisor agent, can improve the system’s efficiency and can scale to handle large volumes of data or complex research queries without bottlenecking at any single point of the process.

[0050] In some implementations, the supervisor agent can be equipped with the capability to set a weight for each agent within the multi-agent scientific research system. For example, the weight may be pre-defined or may be adaptive. The weight may be adaptive based on the number of hypotheses already generated, the number of iterations that have been performed, and / or other factors. This weighting can influence the relative importance or priority of each agent’s tasks in the overall research process. For example, if the hypothesis generation agent’s tasks are deemed more immediately beneficial to a project’s progress, thesupervisor agent can assign a higher weight to this agent, thereby allocating more computational resources or worker device time to hypothesis generation activities.Conversely, if the evolution agent’s role in refining and ranking hypotheses becomes more beneficial at a later stage, the supervisor can adjust the weights accordingly to shift resources to these tasks.

[0051] In some implementations, the supervisor agent can sample agent actions from each agent in the multi-agent scientific research system in proportion to their respective weights. Agents with higher weights can have their tasks executed more frequently or with greater computational resources. For example, during a phase of intensive hypothesis testing and validation, the supervisor agent can increase the weight of the evolution agent, which can result in more frequent sampling of its actions such as pairwise comparisons of hypotheses. This dynamic allocation of resources based on task importance and system needs can improve workflow efficiency.

[0052] The systems and methods of the present disclosure provide a number of technical effects and benefits. As one example, the proposed multi-agent scientific research system leverages advanced machine learning models to automate and optimize the generation and evaluation of research hypotheses. This system is designed to handle complex data processing tasks that are inherent in scientific research, such as natural language processing of research goals, generating multiple hypotheses, and evaluating these hypotheses against extensive scientific literature and databases. This not only enhances the efficiency of generating hypotheses but also ensures that they are robust and relevant, thereby improving the overall quality of scientific inquiry and experimentation, which represents and facilitates progress in the scientific arts.

[0053] As another example, the tournament-based evaluation approach utilized in the multi-agent scientific research system significantly reduces computational expenditure by performing relative comparisons between hypotheses, rather than evaluating each hypothesis in isolation. Traditional systems might individually assess each hypothesis against a set of criteria, which can be computationally intensive especially with large sets of hypotheses. In contrast, the tournament-based method streamlines this process by comparing pairs of hypotheses to determine which one is more promising based on predefined criteria. This pairwise comparison reduces the number of evaluations needed, as less promising hypotheses are quickly filtered out in initial rounds, focusing computational resources on refining and evaluating only the most promising hypotheses. Consequently, this approach not onlyaccelerates the hypothesis ranking process but also optimizes the use of computational resources by minimizing unnecessary evaluations.

[0054] As another example, the process of iteratively modifying hypotheses based on lessons learned or identified relative to other hypotheses significantly enhances the efficiency and effectiveness of scientific hypothesis generation. By employing a pairwise ranking process and a multi-turn self-play or debate mechanism, the system can quickly identify strengths and weaknesses of each hypothesis. This allows for the continuous refinement of hypotheses by incorporating insights and lessons learned from the comparative evaluation of other hypotheses. As a result, the system can more rapidly converge on an optimal hypothesis that is well-founded and robust, thereby reducing computational expenditure.

[0055] In some implementations, the hypothesis generation agent can automatically generate multiple research hypothesis statements in parallel. This parallel processing capability can improve efficiency in hypothesis generation, allowing the system to manage large datasets and address complex scientific questions more effectively.

[0056] In some implementations, the evolution agent can perform the tournamentbased evaluation by executing multiple pairwise evaluations in parallel. This method can improve the speed and efficiency of the evaluation process by enabling simultaneous comparisons of multiple pairs of hypotheses. For instance, in situations where a large number of hypotheses require ranking based on their scientific validity and relevance, the evolution agent can evaluate different pairs of hypotheses concurrently, thereby facilitating the process of identifying the most promising hypotheses for further investigation.

[0057] In some implementations, additional computational aspects of the technology can include the optimization of data handling and processing speeds. For example, the system can leverage advanced data structuring and indexing techniques that facilitate the rapid retrieval and processing of large scientific datasets. This can include the use of parallel computing architectures where data is distributed across multiple processing units, allowing simultaneous access and processing that may improve the efficiency of data-intensive tasks such as hypothesis generation and evaluation.

[0058] Additionally, the system can incorporate machine learning optimization algorithms that may improve the efficiency of model training and inference processes. In some implementations, techniques such as batch normalization and adaptive learning rate adjustments can be employed to enhance the convergence speed of the machine learning models utilized in the hypothesis generation and evolution agents. This can not onlyaccelerate the model training phase but also enhance the responsiveness of the system during the hypothesis evaluation phase, enabling real-time processing and ranking of hypotheses.

[0059] Additionally, the system can utilize caching mechanisms to store intermediate results of hypothesis evaluations. This method can reduce the redundant processing of similar hypothesis comparisons, thereby conserving computational resources and potentially improving the overall throughput of the system. For example, if certain pairs of hypotheses have been previously evaluated, the system can retrieve the stored outcomes instead of recomputing the evaluations. This feature can be particularly beneficial in scenarios where hypotheses are incrementally refined and re-evaluated multiple times.

[0060] Various example implementations are described herein with respect to the accompanying Figures.

[0061] Figure 1 illustrates a flowchart of an example research process according to example implementations of aspects of the present disclosure. Research goal 102 represents the initial input for the system, wherein a specific scientific inquiry or objective can be defined. This can include, for instance, a designated area of study or a particular scientific question that requires addressing.

[0062] Planning stage 104 follows the input of the research goal 102. This stage can include parsing and query rewriting, which can assist in refining the research questions and identifying the relevant search space for literature and data pertinent to the research goal 102. In some implementations, activities in this stage can include the identification of keywords, relevant databases, and potential sources of information.

[0063] Generation stage 106 is designed for the formulation of novel hypotheses. This stage can utilize inputs from planning 104 to conduct literature reviews and summarize existing knowledge. Furthermore, generation 106 can propose novel hypotheses based on the synthesized information. This process can include the application of advanced data analytics and machine learning techniques. As one example, the generation stage 106 can be performed by a hypothesis statement generation agent.

[0064] Reflection stage 108 can incorporate expert and real -world feedback into the hypotheses generated in generation 106. This stage can also include deep review and verification processes, wherein the proposed hypotheses can be rigorously tested against existing data and scientific principles to assess their validity and potential impact. As one example, the reflection stage 108 can be performed by a reflection agent.

[0065] Evolution stage 110 can include the refinement of hypotheses through tournaments where hypotheses are evaluated against one another to ascertain their relativemerit. This stage can improve the prioritization of hypotheses that are more likely to yield significant scientific insights. As one example, the evolution stage 110 can be performed by an evolution agent.

[0066] Research plan 112 represents the final output of the system, in which a comprehensive plan for investigating the selected hypotheses is formulated. This plan can detail the experimental designs, methodologies, and / or resources that can be utilized to test the hypotheses and improve the likelihood of achieving research goal 102.

[0067] Figure 2 illustrates a workflow diagram for a multi-agent scientific research system 202 designed to facilitate the generation, evaluation, and refinement of research hypotheses. As illustrated by figure 2, a human scientist 204 can initiate the research process by supplying a research goal and experimental constraints. This is represented by the section labeled “start research, provide a goal and experiment constraints 205a.” In some implementations, the user can also be provided with the ability to add an idea 205b, review an idea 205c, and / or discuss the research 205d.

[0068] The multi-agent system 202 can generate a configuration plan 206 in which research objectives and preferences can be established. This configuration plan 206 can guide the generation and refinement of hypotheses during the research process.

[0069] In a generate ideas workflow 208a, the system 202 can automatically generate multiple research hypotheses based on input from the configuration 206. This process can include exploring literature 208b and engaging in scientific debates 208e to improve the novelty and relevance of the hypotheses. These hypotheses can then be ranked and stored, as represented by box 207. As one example, the generate ideas workflow 208a can be performed by a hypothesis statement generation agent.

[0070] The review ideas workflow 210a can include the evaluation of generated hypotheses by assessing attributes such as correctness 210b and novel explanations 210c. In some implementations, common weaknesses detected in reviews 210c can be identified, which can be beneficial for enhancing the robustness of the hypotheses. As one example, the review ideas workflow 210a can be performed by a reflection agent.

[0071] In a compare ideas workflow 212a, hypotheses can be compared, typically including further scientific debate as shown at 212b. The system 202 can identify the types of top wins and losses in section 212c, which can assist in refining the hypotheses further. As one example, the compare ideas workflow 212a can be performed by an evolution agent.

[0072] In an improve ideas workflow 214a, enhancements to hypotheses can be performed. This can include deriving inspiration from multiple ideas 214b, simplifyingexisting hypotheses 214c, and / or extending the research 214d to encompass additional or previously neglected areas. As one example, the improve ideas workflow 214a can be performed by an improve agent.

[0073] The research overview workflow 216 can synthesize information and refinements into a coherent overview. In some implementations, the system 202 incorporates functionalities for search 218a and tools 218b, which can facilitate the research process by providing access to external databases, analytical tools, and / or experimental software.

[0074] Figure 3 illustrates a block diagram of an example multi-agent system architecture 302 according to example implementations of aspects of the present disclosure. Research goal 304 in Figure 3 illustrates the initial input for the system, where a specific scientific inquiry or objective is defined. This can be a user-inputted directive that outlines the primary focus of the research, such as investigating a particular scientific phenomenon or developing a new technological application.

[0075] Configuration plan 306 can set parameters and constraints that tailor the research process to specific requirements or conditions. This can include defining the scope of the research, the resources available, and any particular methodologies or data to be utilized. In some implementations, configuration plan 306 can specify the utilization of certain datasets or the exclusion of specific types of information.

[0076] Supervisor agent 308 is tasked with overseeing the operation of the system, which can improve the efficiency of task assignment and execution. The supervisor agent 308 can allocate resources, prioritize tasks, and / or direct the flow of information among various system components. In some implementations, supervisor agent 308 can assign different weightings to various agents depending on the current phase of the research or specific needs that emerge.

[0077] Worker(s) 310 comprises components or agents within the system that can perform specific tasks assigned by supervisor agent 308. These tasks can include data processing, analysis, and hypothesis generation, among other functions relevant to the research process. As an example, in some implementations, worker(s) 310 can be tasked with analyzing a dataset to extract relevant information for hypothesis generation.

[0078] The multi-agent system 302 can implement each of its agents, as described further below, to ultimately generate a research proposal 312. The research proposal 312 can be a final, actionable proposal which can be empirically evaluated. In some implementations, the system 302 can operate iteratively, such that the generated proposal 312 from a first iteration can be provided as input for refinement at a next iteration.

[0079] As illustrated in Figure 3, the multi -agent system 302 can include a number of different agents. An idea generation agent 314 (also known as hypothesis statement generation agent) can include processes for formulating new research hypotheses based on available input parameters and data. This can include the utilization of machine-learned models to synthesize new ideas, or the application of machine-learned models to identify patterns or predictions that can suggest plausible hypotheses.

[0080] A reflection agent 316 can include the review and refinement of hypotheses generated by idea generation 314. This process can include assessing the validity of the hypotheses, comparing them to existing knowledge, and refining the ideas based on feedback and further analysis. In some implementations, reflection agent 316 can utilize a peer review process simulated by machine-learned models to evaluate the soundness of the proposed hypotheses.

[0081] An evolution agent 318 performs a tournament-based ranking of hypotheses. The tournament-based evaluation process implemented by evolution agent 318 can include iterative pairwise comparisons among the generated hypotheses. The process begins with the selection of pairs of hypotheses from the pool of generated hypotheses. Each pair undergoes a comparative evaluation to determine which hypothesis of the pair is more robust or relevant based on predefined evaluation criteria. These criteria can include factors such as novelty, scientific rigor, alignment with current research, empirical support, and potential impact on the field.

[0082] To perform the evaluations, evolution agent 318 can utilize one or more machine-learned models. These models are capable of analyzing the content and context of each hypothesis, comparing them based on the evaluation criteria. After each pairwise comparison, the hypotheses are ranked. The hypothesis that scores higher based on the evaluation criteria is moved up in the ranking. This process is repeated across multiple rounds, with hypotheses being continuously compared and re-ranked against new opponents from the pool.

[0083] An improve agent 320 can include mechanisms that enhance or optimize hypotheses. This can include integrating new data, applying various analytical techniques, or redefining the scope or focus of the hypothesis. For instance, improve agent 320 can utilize feedback from experimental results to adjust the parameters of a hypothesis or to consider alternative explanations.

[0084] A meta-review agent 322 can include a higher-level analysis of the research process and outcomes, assessing the overall effectiveness and efficiency of the researchsystem. This can include evaluating the performance of different components, such as idea generation agent 314 or reflection agent 316. In some implementations, adjustments can be made to enhance future research cycles.

[0085] A proximity graph agent 324 can provide a visual or computational representation of the relationships between various hypotheses or concepts generated during the research process. This can assist in identifying clusters of related ideas and understanding the relationships between different hypotheses, thereby improving the guidance for further exploration or refinement of research ideas. For instance, proximity graph agent 324 can illustrate how different hypotheses are interconnected, which can aid researchers in determining which areas can be beneficial for further investigation.

[0086] Figure 4 depicts a flowchart of a method 400 for training one or more machine-learned models according to aspects of the present disclosure. One or more portion(s) of example method 400 can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of example method 400 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 400 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. Figure 4 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. Figure 4 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of example method 400 can be performed additionally, or alternatively, by other systems.

[0087] At 402, example method 400 can include obtaining a training instance. A set of training data can include a plurality of training instances divided between multiple datasets (e.g., a training dataset, a validation dataset, or testing dataset). A training instance can be labeled or unlabeled. Although referred to in example method 400 as a “training” instance, it is to be understood that runtime inferences can form training instances when a model is trained using an evaluation of the model’s performance on that runtime instance (e.g., online training / learning). Example data types for the training instance and various tasks associated therewith are described throughout the present disclosure.

[0088] At 404, example method 400 can include processing, using one or more machine-learned models, the training instance to generate an output. The output can be directly obtained from the one or more machine-learned models or can be a downstream result of a chain of processing operations that includes an output of the one or more machine-learned models.

[0089] At 406, example method 400 can include receiving an evaluation signal associated with the output. The evaluation signal can be obtained using a loss function.Various determinations of loss can be used, such as mean squared error, likelihood loss, cross entropy loss, hinge loss, contrastive loss, or various other loss functions. The evaluation signal can be computed using known ground-truth labels (e.g., supervised learning), predicted or estimated labels (e.g., semi- or self-supervised learning), or without labels (e.g., unsupervised learning). The evaluation signal can be a reward (e.g., for reinforcement learning). The reward can be computed using a machine-learned reward model configured to generate rewards based on output(s) received. The reward can be computed using feedback data describing human feedback on the output(s).

[0090] At 408, example method 400 can include updating the machine-learned model using the evaluation signal. For example, values for parameters of the machine-learned model(s) can be learned, in some embodiments, using various training or learning techniques, such as, for example, backwards propagation. For example, the evaluation signal can be backpropagated from the output (or another source of the evaluation signal) through the machine-learned model(s) to update one or more parameters of the model(s) (e.g., based on a gradient of the evaluation signal with respect to the parameter value(s)). For example, system(s) containing one or more machine-learned models can be trained in an end-to-end manner. Gradient descent techniques can be used to iteratively update the parameters over a number of training iterations. In some implementations, performing backwards propagation of errors can include performing truncated backpropagation through time. Example method 400 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0091] In some implementations, example method 400 can be implemented for training a machine-learned model from an initialized state to a fully trained state (e.g., when the model exhibits a desired performance profile, such as based on accuracy, precision, recall, etc.).

[0092] In some implementations, example method 400 can be implemented for particular stages of a training procedure. For instance, in some implementations, examplemethod 400 can be implemented for pre-training a machine-learned model. Pre-training can include, for instance, large-scale training over potentially noisy data to achieve a broad base of performance levels across a variety of tasks / data types.

[0093] In some implementations, example method 400 can be implemented for finetuning a machine-learned model. Fine-tuning can include, for instance, smaller-scale training on higher-quality (e.g., labeled, curated, etc.) data. Fine-tuning can affect all or a portion of the parameters of a machine-learned model. For example, various portions of the machine-learned model can be “frozen” for certain training stages. For example, parameters associated with an embedding space can be “frozen” during fine-tuning (e.g., to retain information learned from a broader domain(s) than present in the fine-tuning dataset(s)). In some implementations, example method 400 uses adapter modules. Adapters can be small trainable layers that are inserted between pre-existing layers of a pre-trained model. During the finetuning process, the original parameters of the pre-trained model are typically frozen, and only the parameters of the adapters are updated.

[0094] In some implementations, example method 400 can be implemented to execute parameter-efficient fine-tuning methods, such as Layerwise Optimization of Residuals (LoRA). LoRA can refine pre-trained models with minimal adjustments to the original parameters. This can be achieved by introducing trainable low-rank matrices that modify the behavior of the pre-trained weights without directly altering them. In some implementations, during fine-tuning, only these auxiliary matrices are updated, which significantly reduces the number of parameters that are trained.

[0095] An example fine-tuning approach includes reinforcement learning.Reinforcement learning can be based on user feedback on model performance during use.

[0096] Figure 5 is a block diagram of an example processing flow for using machine-learned model(s) 1 to process input(s) 2 to generate output(s) 3.

[0097] Machine-learned model(s) 1 can be or include one or multiple machine-learned models or model components. Example machine-learned models can include neural networks (e.g., deep neural networks). Example machine-learned models can include nonlinear models or linear models. Example machine-learned models can use other architectures in lieu of or in addition to neural networks. Example machine-learned models can include decision tree based models, support vector machines, hidden Markov models, Bayesian networks, linear regression models, k-means clustering models, etc.

[0098] Machine-learned model(s) 1 can be or include, or otherwise be representative of any one or more of the machine-learned models described above with respect to the1preceding figures. For example, machine-learned model(s) 1 can be or include, or otherwise be representative of any one or more of any of the machine-learned components described herein, etc. Although various features, variations, and implementations described below are described with respect to machine-learned model(s) 1, it is to be understood that such features, variations, and implementations are to be understood as described with respect to each any machine-learned component described herein.

[0099] Example neural networks can include feed-forward neural networks, recurrent neural networks (RNNs), including long short-term memory (LSTM) based recurrent neural networks, convolutional neural networks (CNNs), diffusion models, generative-adversarial networks, or other forms of neural networks. Example neural networks can be deep neural networks. Some example machine-learned models can leverage an attention mechanism such as self-attention. For example, some example machine-learned models can include multiheaded self-attention models.

[0100] Machine-learned model(s) 1 can include a single or multiple instances of the same model configured to operate on data from input(s) 2. Machine-learned model(s) 1 can include multiple different models or multiple different model portions configured to operate on data from input(s) 2.

[0101] Machine-learned model(s) 1 can include an ensemble of different models that can cooperatively interact to process data from input(s) 2. For example, a model ensemble can include multiple models that have different attributes (e.g., different architectures, trained with different recipes, etc.). The ensemble can output an overall output based on the individual outputs of the constituent models. In this manner, for instance, the diverse constituent models can work together to provide system-level robustness by effectively aggregating over individual strengths and weaknesses of any given model. The respective individual outputs can be combined in a weighted combination, using a voting or routing mechanism, or a learned output layer (e.g., one or more feedforward or fully-connected layers).

[0102] Machine-learned model(s) 1 can employ a mixture-of-experts structure. See, e.g., Zhou et al., Mixture-of-Experts with Expert Choice Routing, ARXlV:2202.09368v2 (Oct.14, 2022). For example, different portions of a model can learn (explicitly or implicitly) different expertise areas, with pathways through the model being selected by a learned routing mechanism that engages the appropriate expert for a given input (e.g., a given portion of an input, such as on a per-token basis). For example, a feedforward network can be sparsely activated for a given portion of an input based on an output of a routing mechanismthat processes the portion of the input. In this manner, for instance, the group of activated weights can form an “expert” that is selected by the router. On each forward pass, only a subset of the total model weights can be engaged, thereby decreasing a quantity of operations performed for processing a given input compared to a densely activated model. In this manner, for instance, the expressive and interpretive power of a high-parameter-count model can be achieved with more compute-efficient forward passes.

[0103] Input(s) 2 can generally include or otherwise represent various types of data. Input(s) 2 can include one type or many different types of data. Output(s) 3 can be data of the same type(s) or of different types of data as compared to input(s) 2. Output(s) 3 can include one type or many different types of data.

[0104] Example data types for input(s) 2 or output(s) 3 include natural language text data, software code data (e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages), machine code data (e.g., binary code, assembly code, or other forms of machine-readable instructions that can be executed directly by a computer’s central processing unit), assembly code data (e.g., low-level programming languages that use symbolic representations of machine code instructions to program a processing unit), genetic data or other chemical or biochemical data, image data, audio data, audiovisual data, haptic data, biometric data, medical data, financial data, statistical data, geographical data, astronomical data, historical data, sensor data generally (e.g., digital or analog values, such as voltage or other absolute or relative level measurement values from a real or artificial input, such as from an audio sensor, light sensor, displacement sensor, etc.), and the like. Data can be raw or processed and can be in any format or schema.

[0105] In multimodal inputs 2 or outputs 3, example combinations of data types include image data and audio data, image data and natural language data, natural language data and software code data, image data and biometric data, sensor data and medical data, etc. It is to be understood that any combination of data types in an input 2 or an output 3 can be present.

[0106] An example input 2 can include one or multiple data types, such as the example data types noted above. An example output 3 can include one or multiple data types, such as the example data types noted above. The data type(s) of input 2 can be the same as or different from the data type(s) of output 3. It is to be understood that the example data types noted above are provided for illustrative purposes only. Data types contemplated within the scope of the present disclosure are not limited to those examples noted above.

[0107] Figure 6 is a block diagram of an example implementation of an example machine-learned model configured to process sequences of information. For instance, an example implementation of machine-learned model(s) 1 can include machine-learned sequence processing model(s) 4. An example system can pass input(s) 2 to sequence processing model(s) 4. Sequence processing model(s) 4 can include one or more machine-learned components. Sequence processing model(s) 4 can process the data from input(s) 2 to obtain an input sequence 5. Input sequence 5 can include one or more input elements 5-1, 5-2, . . . , 5-A , etc. obtained from input(s) 2. Sequence processing model 4 can process input sequence 5 using prediction layer(s) 6 to generate an output sequence 7. Output sequence 7 can include one or more output elements 7-1, 7-2, . . . , 7 -TV, etc. generated based on input sequence 5. The system can generate output(s) 3 based on output sequence 7.

[0108] Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences of information. For example, some example sequence processing models are referred to as language models and can leverage language-based understandings across one or multiple modalities of input information. Sequence processing model(s) 4 can include relatively large models (e.g., more parameters, computationally expensive, etc.), which can be referred to as “Large Language Models” or LLMs. Sequence processing model(s) 4 can include relatively small models (e.g., fewer parameters, computationally lightweight, etc.), which can be referred to as “Small Language Models” or SLMs. Example language models include, for instance, models described in Gemma: Open Models Based on Gemini Research and Technology, GOOGLE, https: / / arxiv.org / abs / 2403.08295; Gemma 2: Improving Open Language Models at a Practical Size, GOOGLE, https: / / arxiv.org / abs / 2408.00118.

[0109] Sequence processing model(s) 4 can process one or multiple types of data simultaneously. Variations of language models that can perform joint vision and language tasks can be referred to as “Vision-Language Models,” or VLMs. Example VLMs include models described in PaliGemma: A versatile 3B VLM for transfer, GOOGLE, https: / / arxiv.org / abs / 2407.07726; PaliGemma 2: A Family of Versatile VLMs for Transfer, GOOGLE, https: / / arxiv.org / abs / 2412.03555; Flamingo: a Visual Language Model for FewShot Learning, GOOGLE, https: / / arxiv.org / abs / 2204.14198; PaLI: A Jointly-Scaled Multilingual Language-Image Model, GOOGLE, https: / / arxiv.org / abs / 2209.06794.

[0110] Sequence processing model(s) 4 can be multimodal. Example multimodal sequence processing models include, for instance, models described in Gemini: A Family of Highly Capable Multimodal Models, GOOGLE, https: / / arxiv.org / abs / 2312.11805; Gemini 1.5:Unlocking multimodal understanding across millions of tokens of context, GOOGLE, https: / / arxiv.org / abs / 2403.05530.

[0111] Other example sequence processing models can operate to generate outputs or receive inputs in specific domains, such as image domains, see, e.g., Dosovitskiy et al., An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, ARXlV:2010.11929v2 (Jun. 3, 2021), audio domains, see, e.g., Agostinelli et A., MusicIM: Generating Music From Text, ARXlV:2301.11325vl (Jan. 26, 2023), biochemical domains, see, e.g., Jumper et al., Highly accurate protein structure prediction with AlphaFold, 596 Nature 583 (Aug. 26, 2021), by way of example.

[0112] In general, sequence processing model(s) 4 can obtain input sequence 5 using data from input(s) 2. For instance, input sequence 5 can include a representation of data from input(s) 2 in a format understood by sequence processing model(s) 4. One or more machine-learned components of sequence processing model(s) 4 can ingest the data from input(s) 2, parse the data into pieces compatible with the processing architectures of sequence processing model(s) 4 (e.g., via “tokenization”), and project the pieces into an input space associated with prediction layer(s) 6 (e.g., via “embedding”).

[0113] Sequence processing model(s) 4 can ingest the data from input(s) 2 and parse the data into a sequence of elements to obtain input sequence 5. For example, a portion of input data from input(s) 2 can be broken down into pieces that collectively represent the content of the portion of the input data. The pieces can provide the elements of the sequence.

[0114] Elements 5-1, 5-2, . . . , 5-M can represent, in some cases, building blocks for capturing or expressing meaningful information in a particular data domain. For instance, the elements can describe “atomic units” across one or more domains. For example, for textual input source(s), the elements can correspond to groups of one or more words or sub-word components, such as sets of one or more characters.

[0115] For example, elements 5-1, 5-2, . . . , 5-M can represent tokens obtained using a tokenizer. For instance, a tokenizer can process a given portion of an input source and output a series of tokens (e.g., corresponding to input elements 5-1, 5-2, . . . , 5-M) that represent the portion of the input source. Various approaches to tokenization can be used. For instance, textual input source(s) can be tokenized using a byte-pair encoding (BPE) technique. See, e.g., Kudo et al., SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing, PROCEEDINGS OF THE 2018 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66-71 (October 31-November 4, 2018),https: / / aclanthology.org / D18-2012.pdf. Image-based input source(s) can be tokenized by extracting and serializing patches from an image.

[0116] In general, arbitrary data types can be serialized and processed into input sequence 5. It is to be understood that element(s) 5-1, 5-2, . . . , 5-M depicted in Figure 6 can be the tokens or can be the embedded representations thereof.

[0117] Prediction layer(s) 6 can predict one or more output elements 7-1, 7-2, . . . , 7-N based on the input elements. Prediction layer(s) 6 can include one or more machine-learned model architectures, such as one or more layers of learned parameters that manipulate and transform the input(s) to extract higher-order meaning from, and relationships between, input element(s) 5-1, 5-2, . . . , 5-M. In this manner, for instance, example prediction layer(s) 6 can predict new output element(s) in view of the context provided by input sequence 5.

[0118] Prediction layer(s) 6 can evaluate associations between portions of input sequence 5 and a particular output element. These associations can inform a prediction of the likelihood that a particular output follows the input context. For example, consider the textual snippet, “The carpenter’s toolbox was small and heavy. It was full of .” Example prediction layer(s) 6 can identify that “It” refers back to “toolbox” by determining a relationship between the respective embeddings. Example prediction layer(s) 6 can also link “It” to the attributes of the toolbox, such as “small” and “heavy.” Based on these associations, prediction layer(s) 6 can, for instance, assign a higher probability to the word “nails” than to the word “sawdust.”

[0119] A transformer is an example architecture that can be used in prediction layer(s) 4. See, e.g., Vaswani et al., Attention Is All You Need, ARXlV:1706.03762v7 (Aug. 2, 2023). A transformer is an example of a machine-learned model architecture that uses an attention mechanism to compute associations between items within a context window. The context window can include a sequence that contains input sequence 5 and potentially one or more output element(s) 7-1, 7-2, . . . , 1-N. A transformer block can include one or more attention layer(s) and one or more post-attention layer(s) (e.g., feedforward layer(s), such as a multi-layer perceptron).

[0120] Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memory (LSTM) models can also be used, as well as convolutional neural networks (CNNs). In general, prediction layer(s) 6 can leverage various kinds of artificial neural networks that can understand or generate sequences of information.

[0121] Output sequence 7 can include or otherwise represent the same or different data types as input sequence 5. For instance, input sequence 5 can represent textual data, and output sequence 7 can represent textual data. Input sequence 5 can represent image, audio, or audiovisual data, and output sequence 7 can represent textual data (e.g., describing the image, audio, or audiovisual data). It is to be understood that prediction layer(s) 6, and any other interstitial model components of sequence processing model(s) 4, can be configured to receive a variety of data types in input sequence(s) 5 and output a variety of data types in output sequence(s) 7.

[0122] Output sequence 7 can have various relationships to input sequence 5. Output sequence 7 can be a continuation of input sequence 5. Output sequence 7 can be complementary to input sequence 5. Output sequence 7 can translate, transform, augment, or otherwise modify input sequence 5. Output sequence 7 can answer, evaluate, confirm, or otherwise respond to input sequence 5. Output sequence 7 can implement (or describe instructions for implementing) an instruction provided via input sequence 5.

[0123] Output sequence 7 can be generated autoregressively. For instance, for some applications, an output of one or more prediction layer(s) 6 can be passed through one or more output layers (e.g., softmax layer) to obtain a probability distribution over an output vocabulary (e.g., a textual or symbolic vocabulary) conditioned on a set of input elements in a context window. In this manner, for instance, output sequence 7 can be autoregressively generated by sampling a likely next output element, adding that element to the context window, and re-generating the probability distribution based on the updated context window, and sampling a likely next output element, and so forth.

[0124] Output sequence 7 can also be generated non-autoregressively. For instance, multiple output elements of output sequence 7 can be predicted together without explicit sequential conditioning on each other. See, e.g., Saharia et al., Non-Autoregressive Machine Translation with Latent Alignments, ARXlV:2004.07437v3 (NOV. 16, 2020).

[0125] Output sequence 7 can include one or multiple portions or elements. In an example content generation configuration, output sequence 7 can include multiple elements corresponding to multiple portions of a generated output sequence (e.g., a textual sentence, values of a discretized waveform, computer code, etc.). In an example classification configuration, output sequence 7 can include a single element associated with a classification output. For instance, an output “vocabulary” can include a set of classes into which an input sequence is to be classified. For instance, a vision transformer block can pass latent stateinformation to a multilayer perceptron that outputs a likely class value associated with an input image.

[0126] Figure 7 is a block diagram of an example technique for populating an example input sequence 8. Input sequence 8 can include various functional elements that form part of the model infrastructure, such as an element 8-0 obtained from a task indicator 9 that signals to any model(s) that process input sequence 8 that a particular task is being performed (e.g., to help adapt a performance of the model(s) to that particular task). Input sequence 8 can include various data elements from different data modalities. For instance, an input modality 10-1 can include one modality of data. A data-to-sequence model 11-1 can process data from input modality 10-1 to project the data into a format compatible with input sequence 8 (e.g., one or more vectors dimensioned according to the dimensions of input sequence 8) to obtain elements 8-1, 8-2, 8-3. Another input modality 10-2 can include a different modality of data. A data-to-sequence model 11-2 can project data from input modality 10-2 into a format compatible with input sequence 8 to obtain elements 8-4, 8-5, 8-6. Another input modality 10-3 can include yet another different modality of data. A data-to-sequence model 11-3 can project data from input modality 10-3 into a format compatible with input sequence 8 to obtain elements 8-7, 8-8, 8-9.

[0127] Input sequence 8 can be the same as or different from input sequence 5. Input sequence 8 can be a multimodal input sequence that contains elements that represent data from different modalities using a common dimensional representation. For instance, an embedding space can have P dimensions. Input sequence 8 can be configured to contain a plurality of elements that have dimensions. In this manner, for instance, example implementations can facilitate information extraction and reasoning across diverse data modalities by projecting data into elements in the same embedding space for comparison, combination, or other computations therebetween.

[0128] For example, elements 8-0, . . . , 8-9 can indicate particular locations within a multidimensional embedding space. Some elements can map to a set of discrete locations in the embedding space. For instance, elements that correspond to discrete members of a predetermined vocabulary of tokens can map to discrete locations in the embedding space that are associated with those tokens. Other elements can be continuously distributed across the embedding space. For instance, some data types can be broken down into continuously defined portions (e.g., image patches) that can be described using continuously distributed locations within the embedding space.

[0129] In some implementations, the expressive power of the embedding space can not be limited to meanings associated with any particular set of tokens or other building blocks. For example, a continuous embedding space can encode a spectrum of high-order information. An individual piece of information (e.g., a token) can map to a particular point in that space: for instance, a token for the word “dog” can be projected to an embedded value that points to a particular location in the embedding space associated with canine-related information. Similarly, an image patch of an image of a dog on grass can also be projected into the embedding space. In some implementations, the projection of the image of the dog can be similar to the projection of the word “dog” while also having similarity to a projection of the word “grass,” while potentially being different from both. In some implementations, the projection of the image patch can not exactly align with any single projection of a single word. In some implementations, the projection of the image patch can align with a combination of the projections of the words “dog” and “grass.” In this manner, for instance, a high-order embedding space can encode information that can be independent of data modalities in which the information is expressed.

[0130] Task indicator 9 can include a model or model component configured to identify a task being performed and inject, into input sequence 8, an input value represented by element 8-0 that signals which task is being performed. For instance, the input value can be provided as a data type associated with an input modality and projected along with that input modality (e.g., the input value can be a textual task label that is embedded along with other textual data in the input; the input value can be a pixel-based representation of a task that is embedded along with other image data in the input; etc.). The input value can be provided as a data type that differs from or is at least independent from other input(s). For instance, the input value represented by element 8-0 can be learned within a continuous embedding space.

[0131] Input modalities 10-1, 10-2, and 10-3 can be associated with various different data types (e.g., as described above with respect to input(s) 2 and output(s) 3).

[0132] Data-to-sequence models 11-1, 11-2, and 11-3 can be the same or different from each other. Data-to-sequence models 11-1, 11-2, and 11-3 can be adapted to each respective input modality 10-1, 10-2, and 10-3. For example, a textual data-to-sequence model can subdivide a portion of input text and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-1, 8-2, 8-3, etc.). An image data-to-sequence model can subdivide an input image and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-4, 8-5, 8-6, etc.). An arbitrary data type data-to-sequence model cansubdivide an input of that arbitrary data type and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-7, 8-8, 8-9, etc.).

[0133] Data-to-sequence models 11-1, 11-2, and 11-3 can form part of machine-learned sequence processing model(s) 4. Data-to-sequence models 11-1, 11-2, and 11-3 can be jointly trained with or trained independently from machine-learned sequence processing model(s) 4. Data-to-sequence models 11-1, 11-2, and 11-3 can be trained end-to-end with machine-learned sequence processing model(s) 4.

[0134] Figure 8 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.). Model development platform 12 can provide a number of different toolkits that developer systems can employ in the development of new or adapted machine-learned models.

[0135] Model development platform 12 can provide one or more model libraries 13 containing building blocks for new models. Model libraries 13 can include one or more pretrained foundational models 13-1, which can provide a backbone of processing power across various tasks. Model libraries 13 can include one or more pre-trained expert models 13-2, which can be focused on performance in particular domains of expertise. Model libraries 13 can include various model primitives 13-3, which can provide low-level architectures or components (optionally pre-trained), which can be assembled in various arrangements as desired. Model primitives 13-3 can include a library of pre-trained adapters or LoRA modules that can adapt a baseline foundational model to align its outputs with a desired performance profile, augment model capabilities (e.g., to adapt to a different input modality, etc.), and the like.

[0136] Model development platform 12 can receive selections of various model components 14. Model development platform 12 can pass selected model components 14 to a workbench 15 that combines selected model components 14 into a development model 16.

[0137] Workbench 15 can facilitate further refinement and adaptation of development model 16 by leveraging a number of different toolkits integrated with model development platform 12. For example, workbench 15 can facilitate alignment of the development model 16 with a desired performance profile on various tasks using a model alignment toolkit 17.

[0138] Model alignment toolkit 17 can provide a number of tools for causing development model 16 to generate outputs aligned with desired behavioral characteristics. Alignment can include increasing the accuracy, precision, recall, etc. of model outputs.Alignment can include enforcing output styles, schema, or other preferential characteristics ofmodel outputs. Alignment can be general or domain-specific. For instance, a pre-trained foundational model 13-1 can begin with an initial level of performance across multiple domains. Alignment of the pre-trained foundational model 13-1 can include improving a performance in a particular domain of information or tasks (e.g., even at the expense of performance in another domain of information or tasks).

[0139] Model alignment toolkit 17 can integrate one or more dataset(s) 17-1 for aligning development model 16. Curated dataset(s) 17-1 can include labeled or unlabeled training data. Dataset(s) 17-1 can be obtained from public domain datasets. Dataset(s) 17-1 can be obtained from private datasets associated with one or more developer system(s) for the alignment of bespoke machine-learned model(s) customized for private use-cases.

[0140] Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e.g., denoising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre-training pipelines 17-2 can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training. Workbench 15 can implement a pre-training pipeline 17-2 to pre-train development model 16.

[0141] Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher-quality data. Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1. Fine-tuning pipelines 17-3 can update development model 16 by conducting reinforcement learning using reward signals from user feedback signals. Workbench 15 can implement a fine-tuning pipeline 17-3 to finetune development model 16.

[0142] Prompt libraries 17-4 can include sets of inputs configured to induce behavior aligned with desired performance criteria. Prompt libraries 17-4 can include few-shot prompts (e.g., inputs providing examples of desired model outputs for prepending to a desired runtime query), chain-of-thought prompts (e.g., inputs providing step-by-step reasoning within the exemplars to facilitate thorough reasoning by the model), and the like.

[0143] Example prompts can be retrieved from an available repository of prompt libraries 17-4. Example prompts can be contributed by one or more developer systems using workbench 15.

[0144] In some implementations, pre-trained or fine-tuned models can achieve satisfactory performance without exemplars in the inputs. For instance, zero-shot prompts caninclude inputs that lack exemplars. Zero-shot prompts can be within a domain within a training dataset or outside of the training domain(s).

[0145] Prompt libraries 17-4 can include one or more prompt engineering tools. Prompt engineering tools can provide workflows for retrieving or learning optimized prompt values. Prompt engineering tools can facilitate directly learning prompt values (e.g., input element values) based on one or more training iterations. Workbench 15 can implement prompt engineering tools in development model 16.

[0146] Prompt libraries 17-4 can include pipelines for prompt generation. For example, inputs can be generated using development model 16 itself or other machine-learned models. In this manner, for instance, a first model can process information about a task and output an input for a second model to process in order to perform a step of the task. The second model can be the same as or different from the first model. Workbench 15 can implement prompt generation pipelines in development model 16.

[0147] Prompt libraries 17-4 can include pipelines for context injection. For instance, a performance of development model 16 on a particular task can improve if provided with additional context for performing the task. Prompt libraries 17-4 can include software components configured to identify desired context, retrieve the context from an external source (e.g., a database, a sensor, etc.), and add the context to the input prompt. Workbench 15 can implement context injection pipelines in development model 16.

[0148] Although various training examples described herein with respect to model development platform 12 refer to “pre-training” and “fine-tuning,” it is to be understood that model alignment toolkit 17 can generally support a wide variety of training techniques adapted for training a wide variety of machine-learned models. Example training techniques can correspond to the example training method 400 described above.

[0149] Model development platform 12 can include a model plugin toolkit 18. Model plugin toolkit 18 can include a variety of tools configured for augmenting the functionality of a machine-learned model by integrating the machine-learned model with other systems, devices, and software components. For instance, a machine-learned model can use tools to increase performance quality where appropriate. For instance, deterministic tasks can be offloaded to dedicated tools in lieu of probabilistically performing the task with an increased risk of error. For instance, instead of autoregressively predicting the solution to a system of equations, a machine-learned model can recognize a tool to call for obtaining the solution and pass the system of equations to the appropriate tool. The tool can be a traditional system of equations solver that can operate deterministically to resolve the system of equations. Theoutput of the tool can be returned in response to the original query. In this manner, tool use can allow some example models to focus on the strengths of machine-learned models — e.g., understanding an intent in an unstructured request for a task — while augmenting the performance of the model by offloading certain tasks to a more focused tool for rote application of deterministic algorithms to a well-defined problem.

[0150] Model plugin toolkit 18 can include validation tools 18-1. Validation tools 18-1 can include tools that can parse and confirm output(s) of a machine-learned model.Validation tools 18-1 can include engineered heuristics that establish certain thresholds applied to model outputs. For example, validation tools 18-1 can ground the outputs of machine-learned models to structured data sources (e.g., to mitigate “hallucinations”).

[0151] Model plugin toolkit 18 can include tooling packages 18-2 for implementing one or more tools that can include scripts or other executable code that can be executed alongside development model 16. Tooling packages 18-2 can include one or more inputs configured to cause machine-learned model(s) to implement the tools (e.g., few-shot prompts that induce a model to output tool calls in the proper syntax, etc.). Tooling packages 18-2 can include, for instance, fine-tuning training data for training a model to use a tool.

[0152] Model plugin toolkit 18 can include interfaces for calling external application programming interfaces (APIs) 18-3. For instance, in addition to or in lieu of implementing tool calls or tool code directly with development model 16, development model 16 can be aligned to output instructions that initiate API calls to send or obtain data via external systems.

[0153] Model plugin toolkit 18 can integrate with prompt libraries 17-4 to build a catalog of available tools for use with development model 16. For instance, a model can receive, in an input, a catalog of available tools, and the model can generate an output that selects a tool from the available tools and initiates a tool call for using the tool.

[0154] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. For instance, tools for model compression 19-1 can allow development model 16 to be reduced in size while maintaining a desired level of performance. For instance, model compression 19-1 can include quantization workflows, weight pruning and sparsification techniques, etc. Tools for hardware acceleration 19-2 can facilitate the configuration of the model storage and execution formats to operate optimally on different hardware resources. For instance, hardware acceleration 19-2 can include tools for optimally sharding models for distributed processing over multiple processing units for increased bandwidth, lower unified memoryrequirements, etc. Tools for distillation 19-3 can provide for the training of lighter-weight models based on the knowledge encoded in development model 16. For instance, development model 16 can be a highly performant, large machine-learned model optimized using model development platform 12. To obtain a lightweight model for running in resource-constrained environments, a smaller model can be a “student model” that learns to imitate development model 16 as a “teacher model.” In this manner, for instance, the investment in learning the parameters and configurations of development model 16 can be efficiently transferred to a smaller model for more efficient inference.

[0155] Workbench 15 can implement one, multiple, or none of the toolkits implemented in model development platform 12. Workbench 15 can output an output model 20 based on development model 16. Output model 20 can be a deployment version of development model 16. Output model 20 can be a development or training checkpoint of development model 16. Output model 20 can be a distilled, compressed, or otherwise optimized version of development model 16.

[0156] Figure 9 is a block diagram of an example training flow for training a machine-learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of the example training flow can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example training flow can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. Figure 9 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. Figure 9 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of the example training flow can be performed additionally, or alternatively, by other systems.

[0157] Initially, development model 16 can persist in an initial state as an initialized model 21. Development model 16 can be initialized with weight values. Initial weight values can be random or based on an initialization schema. Initial weight values can be based on prior pre-training for the same or for a different model.

[0158] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pretraining stage 22 can be implemented using one or more pre-training pipelines 17-2 over data from dataset(s) 17-1. Pre-training can be omitted, for example, if initialized model 21 is already pre-trained (e.g., development model 16 contains, is, or is based on a pre-trained foundational model or an expert model).

[0159] Pre-trained model 23 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Pre-trained model 23 can be the initial state if development model 16 was already pre-trained. Pre-trained model 23 can undergo fine-tuning in a fine-tuning stage 24. Fine-tuning stage 24 can be implemented using one or more fine-tuning pipelines 17-3 over data from dataset(s) 17-1. Fine-tuning can be omitted, for example, if a pre-trained model has satisfactory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0160] Fine-tuned model 29 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Fine-tuned model 29 can be the initial state if development model 16 was already fine-tuned. Fine-tuned model 29 can undergo refinement with user feedback 26. For instance, refinement with user feedback 26 can include reinforcement learning, optionally based on human feedback from human users of fine-tuned model 25. As reinforcement learning can be a form of fine-tuning, it is to be understood that fine-tuning stage 24 can subsume the stage for refining with user feedback 26. Refinement with user feedback 26 can produce a refined model 27. Refined model 27 can be output to downstream system(s) 28 for deployment or further development.

[0161] In some implementations, computational optimization operations can be applied before, during, or after each stage. For instance, initialized model 21 can undergo computational optimization 29-1 (e.g., using computational optimization toolkit 19) before pre-training stage 22. Pre-trained model 23 can undergo computational optimization 29-2 (e.g., using computational optimization toolkit 19) before fine-tuning stage 24. Fine-tuned model 25 can undergo computational optimization 29-3 (e.g., using computational optimization toolkit 19) before refinement with user feedback 26. Refined model 27 can undergo computational optimization 29-4 (e.g., using computational optimization toolkit 19) before output to downstream system(s) 28. Computational optimization(s) 29-1, . . . , 29-4 can all be the same, all be different, or include at least some different optimization techniques.

[0162] Figure 10 is a block diagram of an inference system for operating one or more machine-learned model(s) 1 to perform inference (e.g., for training, for deployment, etc.). Amodel host 31 can receive machine-learned model(s) 1. Model host 31 can host one or more model instance(s) 31-1, which can be one or multiple instances of one or multiple models. Model host 31 can host model instance(s) 31-1 using available compute resources 31-2 associated with model host 31.

[0163] Model host 31 can perform inference on behalf of one or more client(s) 32. Client(s) 32 can transmit an input request 33 to model host 31. Using input request 33, model host 31 can obtain input(s) 2 for input to machine-learned model(s) 1. Machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3. Using output(s) 3, model host 31 can return an output payload 34 for responding to input request 33 from client(s) 32. Output payload 34 can include or be based on output(s) 3.

[0164] Model host 31 can leverage various other resources and tools to augment the inference task. For instance, model host 31 can communicate with tool interfaces 35 to facilitate tool use by model instance(s) 31-1. Tool interfaces 35 can include local or remote APIs. Tool interfaces 35 can include integrated scripts or other software functionality. Model host 31 can engage online learning interface(s) 36 to facilitate ongoing improvements to machine-learned model(s) 1. For instance, online learning interface(s) 36 can be used within reinforcement learning loops to retrieve user feedback on inferences served by model host 31. Model host 31 can access runtime data source(s) 37 for augmenting input(s) 2 with additional contextual information. For instance, runtime data source(s) 37 can include a knowledge graph 37-1 that facilitates structured information retrieval for information associated with input request(s) 33 (e.g., a search engine service). Runtime data source(s) 37 can include public or private, external or local database(s) 37-2 that can store information associated with input request(s) 33 for augmenting input(s) 2. Runtime data source(s) 37 can include account data 37-3 which can be retrieved in association with a user account corresponding to a client 32 for customizing the behavior of model host 31 accordingly.

[0165] Model host 31 can be implemented by one or multiple computing devices or systems. Client(s) 2 can be implemented by one or multiple computing devices or systems, which can include computing devices or systems shared with model host 31.

[0166] For example, model host 31 can operate on a server system that provides a machine-learning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-area network). Client device(s) can be end-user devices used by individuals. Client device(s) can be server systems that operate client(s) 32 to provide various functionality as a service to downstream end-user devices.

[0167] In some implementations, model host 31 can operate on the same device or system as client(s) 32. Model host 31 can be a machine-learning service that runs on-device to provide machine-learning functionality to one or multiple applications operating on a client device, which can include an application implementing client(s) 32. Model host 31 can be a part of the same application as client(s) 32. For instance, model host 31 can be a subroutine or method implemented by one part of an application, and client(s) 32 can be another subroutine or method that engages model host 31 to perform inference functions within the application. It is to be understood that model host 31 and client(s) 32 can have various different configurations.

[0168] Model instance(s) 31-1 can include one or more machine-learned models that are available for performing inference. Model instance(s) 31-1 can include weights or other model components that are stored in persistent storage, temporarily cached, or loaded into high-speed memory. Model instance(s) 31-1 can include multiple instance(s) of the same model (e.g., for parallel execution of more requests on the same model). Model instance(s) 31-1 can include instance(s) of different model(s). Model instance(s) 31-1 can include cached intermediate states of active or inactive model(s) used to accelerate inference of those models. For instance, an inference session with a particular model can generate significant amounts of computational results that can be re-used for future inference runs (e.g., using a KV cache for transformer-based models). These computational results can be saved in association with that inference session so that session can be executed more efficiently when resumed.

[0169] Compute resource(s) 31-2 can include one or more processors (central processing units, graphical processing units, tensor processing units, machine-learning accelerators, etc.) connected to one or more memory devices. Compute resource(s) 31-2 can include a dynamic pool of available resources shared with other processes. Compute resource(s) 31-2 can include memory devices large enough to fit an entire model instance in a single memory instance. Compute resource(s) 31-2 can also shard model instance(s) across multiple memory devices (e.g., using data parallelization or tensor parallelization, etc.). This can be done to increase parallelization or to execute a large model using multiple memory devices which individually might not be able to fit the entire model into memory.

[0170] Input request 33 can include data for input(s) 2. Model host 31 can process input request 33 to obtain input(s) 2. Input(s) 2 can be obtained directly from input request 33 or can be retrieved using input request 33. Input request 33 can be submitted to model host 31 via an API.

[0171] Model host 31 can perform inference over batches of input requests 33 in parallel. For instance, a model instance 31-1 can be configured with an input structure that has a batch dimension. Separate input(s) 2 can be distributed across the batch dimension (e.g., rows of an array). The separate input(s) 2 can include completely different contexts. The separate input(s) 2 can be multiple inference steps of the same task. The separate input(s) 2 can be staggered in an input structure, such that any given inference cycle can be operating on different portions of the respective input(s) 2. In this manner, for instance, model host 31 can perform inference on the batch in parallel, such that output(s) 3 can also contain the batch dimension and return the inference results for the batched input(s) 2 in parallel. In this manner, for instance, batches of input request(s) 33 can be processed in parallel for higher throughput of output payload(s) 34.

[0172] Output payload 34 can include or be based on output(s) 3 from machine-learned model(s) 1. Model host 31 can process output(s) 3 to obtain output payload 34. This can include chaining multiple rounds of inference (e.g., iteratively, recursively, across the same model(s) or different model(s)) to arrive at a final output for a task to be returned in output payload 34. Output payload 34 can be transmitted to client(s) 32 via an API.

[0173] Online learning interface(s) 36 can facilitate reinforcement learning of machine-learned model(s) 1. Online learning interface(s) 36 can facilitate reinforcement learning with human feedback (RLHF). Online learning interface(s) 36 can facilitate federated learning of machine-learned model(s) 1.

[0174] Model host 31 can access a library of pre-trained adapters or LoRA modules that can adapt a baseline model to align its outputs with a desired performance profile, augment model capabilities (e.g., to adapt to a different input modality, etc.), and the like. For instance, model host 31 can receive an input request to load a customized model, and model host 31 can retrieve one or more components to adapt a baseline model to the custom profile. Model host 31 can determine that a particular functionality is needed for a particular task (e.g., based on an output of a model that preprocesses an input) and retrieve a pre-trained component accordingly.

[0175] Model host 31 can execute machine-learned model(s) 1 to perform inference for various tasks using various types of data. For example, various different input(s) 2 and output(s) 3 can be used for various different tasks. In some implementations, input(s) 2 can be or otherwise represent image data. Machine-learned model(s) 1 can process the image data to generate an output. As an example, machine-learned model(s) 1 can process the image data to generate an image recognition output (e.g., a recognition of the image data, a latentembedding of the image data, an encoded representation of the image data, a hash of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an image segmentation output. As another example, machine-learned model(s) 1 can process the image data to generate an image classification output. As another example, machine-learned model(s) 1 can process the image data to generate an image data modification output (e.g., an alteration of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an encoded image data output (e.g., an encoded and / or compressed representation of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an upscaled image data output. As another example, machine-learned model(s) 1 can process the image data to generate a prediction output.

[0176] In some implementations, the task is a computer vision task. In some cases, input(s) 2 includes pixel data for one or more images and the task is an image processing task. For example, the image processing task can be image classification, where the output is a set of scores, each score corresponding to a different object class and representing the likelihood that the one or more images depict an object belonging to the object class. The image processing task can be object detection, where the image processing output identifies one or more regions in the one or more images and, for each region, a likelihood that region depicts an object of interest. As another example, the image processing task can be image segmentation, where the image processing output defines, for each pixel in the one or more images, a respective likelihood for each category in a predetermined set of categories. For example, the set of categories can be foreground and background. As another example, the set of categories can be object classes. As another example, the image processing task can be depth estimation, where the image processing output defines, for each pixel in the one or more images, a respective depth value. As another example, the image processing task can be motion estimation, where the network input includes multiple images, and the image processing output defines, for each pixel of one of the input images, a motion of the scene depicted at the pixel between the images in the network input.

[0177] In some implementations, input(s) 2 can be or otherwise represent natural language data. Machine-learned model(s) 1 can process the natural language data to generate an output. As an example, machine-learned model(s) 1 can process the natural language data to generate a language encoding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a latent text embedding output. As another example, machine-learned model(s) 1 can process the natural language data to generate atranslation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a classification output. As another example, machine-learned model(s) 1 can process the natural language data to generate a textual segmentation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a semantic intent output. As another example, machine-learned model(s) 1 can process the natural language data to generate an upscaled text or natural language output (e.g., text or natural language data that is higher quality than the input text or natural language, etc.). As another example, machine-learned model(s) 1 can process the natural language data to generate a prediction output (e.g., one or more predicted next portions of natural language content).

[0178] In some implementations, input(s) 2 can be or otherwise represent speech data (e.g., data describing spoken natural language, such as audio data, textual data, etc.).Machine-learned model(s) 1 can process the speech data to generate an output. As an example, machine-learned model(s) 1 can process the speech data to generate a speech recognition output. As another example, machine-learned model(s) 1 can process the speech data to generate a speech translation output. As another example, machine-learned model(s) 1 can process the speech data to generate a latent embedding output. As another example, machine-learned model(s) 1 can process the speech data to generate an encoded speech output (e.g., an encoded and / or compressed representation of the speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate an upscaled speech output (e.g., speech data that is higher quality than the input speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate a textual representation output (e.g., a textual representation of the input speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate a prediction output.

[0179] In some implementations, input(s) 2 can be or otherwise represent latent encoding data (e.g., a latent space representation of an input, etc.). Machine-learned model(s) 1 can process the latent encoding data to generate an output. As an example, machine-learned model(s) 1 can process the latent encoding data to generate a recognition output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a reconstruction output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a search output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a reclustering output. Asanother example, machine-learned model(s) 1 can process the latent encoding data to generate a prediction output.

[0180] In some implementations, input(s) 2 can be or otherwise represent statistical data. Statistical data can be, represent, or otherwise include data computed and / or calculated from some other data source. Machine-learned model(s) 1 can process the statistical data to generate an output. As an example, machine-learned model(s) 1 can process the statistical data to generate a recognition output. As another example, machine-learned model(s) 1 can process the statistical data to generate a prediction output. As another example, machine-learned model(s) 1 can process the statistical data to generate a classification output. As another example, machine-learned model(s) 1 can process the statistical data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the statistical data to generate a visualization output. As another example, machine-learned model(s) 1 can process the statistical data to generate a diagnostic output.

[0181] In some implementations, input(s) 2 can be or otherwise represent sensor data. Machine-learned model(s) 1 can process the sensor data to generate an output. As an example, machine-learned model(s) 1 can process the sensor data to generate a recognition output. As another example, machine-learned model(s) 1 can process the sensor data to generate a prediction output. As another example, machine-learned model(s) 1 can process the sensor data to generate a classification output. As another example, machine-learned model(s) 1 can process the sensor data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the sensor data to generate a visualization output. As another example, machine-learned model(s) 1 can process the sensor data to generate a diagnostic output. As another example, machine-learned model(s) 1 can process the sensor data to generate a detection output.

[0182] In some implementations, machine-learned model(s) 1 can be configured to perform a task that includes encoding input data for reliable and / or efficient transmission or storage (and / or corresponding decoding). For example, the task can be an audio compression task. The input can include audio data and the output can comprise compressed audio data. In another example, the input includes visual data (e.g. one or more images or videos), the output comprises compressed visual data, and the task is a visual data compression task. In another example, the task can comprise generating an embedding for input data (e.g. input audio or visual data). In some cases, the input includes audio data representing a spoken utterance and the task is a speech recognition task. The output can comprise a text output which is mapped to the spoken utterance. In some cases, the task comprises encrypting ordecrypting input data. In some cases, the task comprises a microprocessor performance task, such as branch prediction or memory address translation.

[0183] In some implementations, the task is a generative task, and machine-learned model(s) 1 can be configured to output content generated in view of input(s) 2. For instance, input(s) 2 can be or otherwise represent data of one or more modalities that encodes context for generating additional content.

[0184] In some implementations, the task can be a text completion task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent textual data and to generate output(s) 3 that represent additional textual data that completes a textual sequence that includes input(s) 2. For instance, machine-learned model(s) 1 can be configured to generate output(s) 3 to complete a sentence, paragraph, or portion of text that follows from a portion of text represented by input(s) 2.

[0185] In some implementations, the task can be an instruction-following task.Machine-learned model(s) 1 can be configured to process input(s) 2 that represent instructions to perform a function and to generate output(s) 3 that advance a goal of satisfying the instruction function (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent data of the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the instructions (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the instructions (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward accomplishing the requested functionality. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of performing a function. Multiple steps can be performed, with a final output being obtained that is responsive to the initial instructions.

[0186] In some implementations, the task can be a question answering task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent a question to answer and to generate output(s) 3 that advance a goal of returning an answer to the question (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent dataof the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward answering the question. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of obtaining an answer to the question (e.g., querying a database, performing a computation, executing a script, etc.). Multiple steps can be performed, with a final output being obtained that is responsive to the question.

[0187] In some implementations, the task can be an image generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of image content. The context can include text data, image data, audio data, etc. Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent image data that depicts imagery related to the context. For instance, machine-learned model(s) 1 can be configured to generate pixel data of an image. Values for channel(s) associated with the pixels in the pixel data can be selected based on the context (e.g., based on a probability determined based on the context).

[0188] In some implementations, the task can be an audio generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of audio content. The context can include text data, image data, audio data, etc. Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent audio data related to the context. For instance, machine-learned model(s) 1 can be configured to generate waveform data in the form of an image (e.g., a spectrogram). Values for channel(s) associated with pixels of the image can be selected based on the context. Machine-learned model(s) 1 can be configured to generate waveform data in the form of a sequence of discrete samples of a continuous waveform. Values of the sequence can be selected based on the context (e.g., based on a probability determined based on the context).

[0189] In some implementations, the task can be a data generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of data (e.g., data from various data domains, such as sensor data, image data, multimodal data, statistical data, etc.). The desired data can be, for instance, synthetic data for training other machine-learned models. The context can include arbitrary data type(s).Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent data that aligns with the desired data. For instance, machine-learned model(s) 1 can be configured to generate data values for populating a dataset. Values for the data object(s) can be selected based on the context (e.g., based on a probability determined based on the context).

[0190] Figure 11 is a block diagram of an example networked computing system that can perform aspects of example implementations of the present disclosure. The system can include a number of computing devices and systems that are communicatively coupled over a network 49. An example computing device 50 is described to provide an example of a computing device that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). An example server computing system 60 is described as an example of a server computing system that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Computing device 50 and server computing system(s) 60 can cooperatively interact (e.g., over network 49) to perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Model development platform system 70 is an example system that can host or serve model development platform(s) 12 for development of machine-learned models. Third-party system(s) 80 are example system(s) with which any of computing device 50, server computing system(s) 60, or model development platform system(s) 70 can interact in the performance of various aspects of the present disclosure (e.g., engaging third-party tools, accessing third-party databases or other resources, etc.).

[0191] Network 49 can be any type of communications network, such as a local area network (e.g., intranet), wide area network (e.g., Internet), or some combination thereof and can include any number of wired or wireless links. In general, communication over network 49 can be carried via any type of wired or wireless connection, using a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), or protection schemes (e.g., VPN, secure HTTP, SSL). Network 49 can also be implemented via a system bus. For instance, one or more devices or systems of Figure 11 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0192] Computing device 50 can be any type of computing device, such as, for example, a personal computing device (e.g., laptop or desktop), a mobile computing device (e.g., smartphone or tablet), a gaming console or controller, a wearable computing device, an embedded computing device, a server computing device, a virtual machine operating on a host device, or any other type of computing device. Computing device 50 can be a client computing device. Computing device 50 can be an end-user computing device. Computing device 50 can be a computing device of a service provided that provides a service to an end user (who can use another computing device to interact with computing device 50).

[0193] Computing device 50 can include one or more processors 51 and a memory 52. Processor(s) 51 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 52 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 52 can store data 53 and instructions 54 which can be executed by processor(s) 51 to cause computing device 50 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein.

[0194] Computing device 50 can also include one or more input components that receive user input. For example, a user input component can be a touch-sensitive component (e.g., a touch-sensitive display screen or a touch pad) that is sensitive to the touch of a user input object (e.g., a finger or a stylus). The touch-sensitive component can serve to implement a virtual keyboard. Other example user input components include a microphone, camera, LIDAR, a physical keyboard or other buttons, or other means by which a user can provide user input.

[0195] Computing device 50 can store or include one or more machine-learned models 55. Machine-learned models 55 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 55 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 55 can be received from server computing system(s) 60, model development platform system 70, third party system(s) 80 (e.g., an application distribution platform), or developed locally on computing device 50. Machine-learned model(s) 55 can be loaded into memory 52 and used or otherwise implemented by processor(s) 51. Computing device 50 can implement multiple parallel instances of machine-learned model(s) 55.

[0196] Server computing system(s) 60 can include one or more processors 61 and a memory 62. Processor(s) 61 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 62 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 62 can store data 63 and instructions 64 which can be executed by processor(s) 61 to cause server computing system(s) 60 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein.

[0197] In some implementations, server computing system 60 includes or is otherwise implemented by one or multiple server computing devices. In instances in which server computing system 60 includes multiple server computing devices, such server computing devices can operate according to sequential computing architectures, parallel computing architectures, or some combination thereof.

[0198] Server computing system 60 can store or otherwise include one or more machine-learned models 65. Machine-learned model(s) 65 can be the same as or different from machine-learned model(s) 55. Machine-learned models 65 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 65 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 65 can be received from computing device 50, model development platform system 70, third party system(s) 80, or developed locally on server computing system(s) 60. Machine-learned model(s) 65 can be loaded into memory 62 and used or otherwise implemented by processor(s) 61. Server computing system(s) 60 can implement multiple parallel instances of machine-learned model(s) 65.

[0199] In an example configuration, machine-learned models 65 can be included in or otherwise stored and implemented by server computing system 60 to establish a client-server relationship with computing device 50 for serving model inferences. For instance, server computing system(s) 60 can implement model host 31 on behalf of client(s) 32 on computing device 50. For instance, machine-learned models 65 can be implemented by server computing system 60 as a portion of a web service (e.g., remote machine-learned model hosting service, such as an online interface for performing machine-learned model operations over a network on server computing system(s) 60). For instance, server computing system(s) 60 can communicate with computing device 50 over a local intranet or internet connection. Forinstance, computing device 50 can be a workstation or endpoint in communication with server computing system(s) 60, with implementation of machine-learned models 65 being managed by server computing system(s) 60 to remotely perform inference (e.g., for runtime or training operations), with output(s) returned (e.g., cast, streamed, etc.) to computing device 50. Machine-learned models 65 can work cooperatively or interoperatively with machine-learned models 55 on computing device 50 to perform various tasks.

[0200] Model development platform system(s) 70 can include one or more processors 71 and a memory 72. Processor(s) 71 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 72 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 72 can store data 73 and instructions 74 which can be executed by processor(s) 71 to cause model development platform system(s) 70 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein with respect to model development platform 12. This and other functionality can be implemented by developer tool(s) 75.

[0201] Third-party system(s) 80 can include one or more processors 81 and a memory 82. Processor(s) 81 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 82 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 82 can store data 83 and instructions 84 which can be executed by processor(s) 81 to cause third-party system(s) 80 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein with respect to tools and other external resources called when training or performing inference with machine-learned model(s) 1, 4, 16, 20, 55, 65, etc. (e.g., third-party resource(s) 85).

[0202] Figure 11 illustrates one example arrangement of computing systems that can be used to implement the present disclosure. Other computing system configurations can be used as well. For example, in some implementations, one or both of computing system 50 orserver computing system(s) 60 can implement all or a portion of the operations of model development platform system 70. For example, computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update / train, or refine machine-learned models 1, 4, 16, 20, 55, 65, etc. using one or more techniques described herein with respect to model alignment toolkit 17. In this manner, for instance, computing system 50 or server computing system(s) 60 can develop, update / train, or refine machine-learned models based on local datasets (e.g., for model personalization / customization, as permitted by user data preference selections).

[0203] Figure 12 is a block diagram of an example computing device 98 that performs according to example embodiments of the present disclosure. Computing device 98 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 98 can include a number of applications (e.g., applications 1 through N). Each application can contain its own machine learning library and machine-learned model(s). For example, each application can include a machine-learned model.Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. As illustrated in Figure 12, each application can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, each application can communicate with each device component using an API (e.g., a public API). In some implementations, the API used by each application is specific to that application.

[0204] Figure 13 is a block diagram of an example computing device 99 that performs according to example embodiments of the present disclosure. Computing device 99 can be the same as or different from computing device 98. Computing device 99 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 99 can include a number of applications (e.g., applications 1 through N). Each application can be in communication with a central intelligence layer. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. In some implementations, each application can communicate with the central intelligence layer (and model(s) stored therein) using an API (e.g., a common API across all applications).

[0205] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in Figure 13, a respective machine-learned model can be provided for each application and managed by the central intelligence layer. In other implementations, two or more applications can share a single machine-learned model. For example, in some implementations, the central intelligence layer can provide a single model for all of the applications. In some implementations, the central intelligence layer is included within or otherwise implemented by an operating system of computing device 99.

[0206] The central intelligence layer can communicate with a central device data layer. The central device data layer can be a centralized repository of data for computing device 99. As illustrated in Figure 13, the central device data layer can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, the central device data layer can communicate with each device component using an API (e.g., a private API).

[0207] The technology discussed herein makes reference to servers, databases, software applications, and other computer-based systems, as well as actions taken and information sent to and from such systems. The inherent flexibility of computer-based systems allows for a great variety of possible configurations, combinations, and divisions of tasks and functionality between and among components. For instance, processes discussed herein can be implemented using a single device or component or multiple devices or components working in combination. Databases and applications can be implemented on a single system or distributed across multiple systems. Distributed components can operate sequentially or in parallel.

[0208] While the present subject matter has been described in detail with respect to various specific example embodiments thereof, each example is provided by way of explanation, not limitation of the disclosure. Those skilled in the art, upon attaining an understanding of the foregoing, can readily produce alterations to, variations of, and equivalents to such embodiments. Accordingly, the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. For instance, features illustrated or described as part of one embodiment can be used with another embodiment to yield a still further embodiment. Thus, it is intended that the present disclosure cover such alterations, variations, and equivalents.

[0209] Aspects of the disclosure have been described in terms of illustrative embodiments thereof. Any and all features in the following claims can be combined or rearranged in any way possible, including combinations of claims not explicitly enumerated in combination together, as the example claim dependencies listed herein should not be read as limiting the scope of possible combinations of features disclosed herein. Accordingly, the scope of the present disclosure is by way of example rather than by way of limitation, and the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. Moreover, terms are described herein using lists of example elements joined by conjunctions such as “and,” “or,” “but,” etc. It should be understood that such conjunctions are provided for explanatory purposes only. Clauses and other sequences of items joined by a particular conjunction such as “or,” for example, can refer to “and / or,” “at least one of’, “any combination of’ example elements listed therein, etc. Terms such as “based on” should be understood as “based at least in part on.”

[0210] The term “can” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X can perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

[0211] The term “can” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X can perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

Claims

WHAT IS CLAIMED IS:

1. A computing system for efficiently performing computer-driven scientific research, the computing system comprising:one or more processors; andone or more non-transitory computer-readable media that collectively store computer-readable instructions that, when executed by the one or more processors, cause the computing system to implement a multi-agent scientific research system, the multi-agent scientific research system comprising:a hypothesis generation agent configured to process data descriptive of a scientific research goal with one or more machine-learned models to automatically generate a plurality of research hypothesis statements expressed in a natural language; andan evolution agent configured to iteratively process at least some of the plurality of research hypothesis statements with one or more machine- learned models to perform a tournament-based evaluation that generates a ranking of the plurality of research hypothesis statements.

2. The computing system of any preceding claim, wherein the multi-agent scientific research system is configured to output a research proposal for the scientific research goal, wherein the research proposal comprises one or more of the plurality of research hypothesis statements that have been selected based at least in part on the ranking of the plurality of research hypothesis statements.

3. The computing system of any preceding claim, wherein the multi-agent scientific research system is configured to:receive textual input descriptive of the scientific research goal; andprocess the textual input with one or more machine-learned models to generate a configuration plan that describes the scientific research goal and one or more of: constraints, considerations, information sources, or criteria for success;wherein the hypothesis statement generation agent is configured to process the configuration plan to generate the plurality of research hypothesis statements.

4. The computing system of any preceding claim, wherein the hypothesis statement generation agent is configured to perform zero-shot prompting to generate a plurality of source hypothesis statements.

5. The computing system of any preceding claim, wherein the hypothesis statement generation agent is configured to perform web search, information retrieval, or other tool use to obtain a plurality of scientific research articles and generate a plurality of source hypothesis statements based on the plurality of scientific research articles.

6. The computing system of any preceding claim, wherein the hypothesis statement generation agent is configured to perform a self-play technique to propose, debate, and / or refine a source hypothesis statement to generate one of the research hypothesis statements.

7. The computing system of any preceding claim, wherein the hypothesis statement generation agent is configured to perform a self-play technique in which the one or more machine-learned models of the hypothesis statement generation agent execute a multi-turn natural language evaluation of a source hypothesis statement to generate one of the research hypothesis statements.

8. The computing system of any preceding claim, wherein the hypothesis statement generation agent is configured to generate one or more intermediate assumptions for inclusion in one or more of the plurality of research hypothesis statements.

9. The computing system of any preceding claim, wherein the multi-agent scientific research system further comprises a reflection agent configured to:perform web search, information retrieval, or other tool use to retrieve one or more academic research articles; andperform, with one or more machine-learned models, model reasoning over the one or more of the plurality of research hypothesis statements in view of the one or more academic research articles.

10. The computing system of claim 9, wherein the model reasoning comprises, for each research hypothesis statement:identifying one or more sets of experimental results or physical observations from the one or more academic research articles; anddetermining a likelihood that the research hypothesis statement comprises a logical cause of the one or more sets of experimental results or physical observations.

11. The computing system of any preceding claim, wherein the multi-agent scientific research system further comprises a reflection agent configured to perform a deep verification process on or more of the plurality of research hypothesis statements using one or more machine-learned models, wherein the deep verification process comprises:identifying one or more assumptions or sub-assumptions contained in or implicit from the research hypothesis statement; andevaluating a correctness of each of the one or more assumptions or subassumptions.

12. The computing system of any preceding claim, wherein the evolution agent performs an iterative pairwise comparison of the plurality of research hypothesis statements.

13. The computing system of any preceding claim, wherein the evolution agent processes two or more of the plurality of research hypothesis statements and a fixed zero-shot prompt with the one or more machine-learned models to generate a comparison of two or more of the plurality of research hypothesis statements.

14. The computing system of any preceding claim, wherein the evolution agent controls the one or more machine-learned models to perform a multi-turn natural language debate about merits of two or more of the plurality of research hypothesis statements.

15. The computing system of any preceding claim, wherein the multi-agent scientific research system further comprises an improve agent configured to process one or more of the plurality of research hypothesis statements with one or more machine-learned models to generate one or more modifications to the one or more of the plurality of research hypothesis statements.

16. The computing system of claim 15, wherein the one or more modifications comprise modifying one of the research hypothesis statements based on one or more other, topranking research hypothesis statements.

17. The computing system of claim 15 or 16, wherein the one or more modifications simplify one or more research hypothesis statements, expand one or more research hypothesis statements, add detail to one or more research hypothesis statements, improve the practicality of one or more research hypothesis statements, improve the grounding of one or more research hypothesis statements, or increase the novelty of one or more research hypothesis statements.

18. The computing system of any preceding claim, wherein the multi-agent scientific research system further comprises a meta-review agent configured to process one or more of the plurality of research hypothesis statements and a plurality of academic research articles with one or more machine-learned models to identify one or more common failure points described in one or more of the academic research articles that are not addressed by the one or more research hypothesis statements.

19. The computing system of any preceding claim, wherein the multi-agent scientific research system further comprises a proximity graph agent configured to build a proximity graph between at least a subset of plurality of research hypothesis statements.

20. The computing system of any preceding claim, wherein the multi-agent scientific research system further comprises a supervisor agent configured to orchestrate worker devices to execute the hypothesis generation agent and the evolution agent.

21. The computing system of claim 20, wherein the supervisor agent is configured to set a weight for each agent in the multi-agent scientific research system and to sample agent actions from each agent in the multi-agent scientific research system in proportion to its respective weight.

22. The computing system of any preceding claim, wherein the hypothesis generation agent is configured to automatically generate the plurality of research hypothesis statements in parallel.

23. The computing system of any preceding claim, wherein the evolution agent is configured to perform the tournament-based evaluation over multiple pairwise evaluations executed in parallel.

24. A computer-implemented method comprising:receiving textual input descriptive of a scientific research goal; andprocessing the textual input with one or more machine-learned models to generate a configuration plan that describes the scientific research goal and one or more of: constraints, considerations, information sources, or criteria for success;processing the configuration plan with one or more machine-learned models to generate a plurality of research hypothesis statements; anditeratively performing a pairwise ranking process, each iteration of the pairwise ranking process comprising:sampling two of the research hypothesis statements; andprocessing the two research hypothesis statements with one or more machine-learned models to select a higher ranked statement and a lower ranked statement.

25. The computer-implemented method of claim 24, wherein processing the configuration plan with one or more machine-learned models to generate a plurality of research hypothesis statements comprises, for each of a plurality of generative iterations: controlling the one or more machine-learned models to perform a multi-turn self-play process that iteratively refines one of the research hypothesis statements.

26. The computer-implemented method of claim 24 or 25, wherein, for each iteration of the pairwise ranking process, processing the two research hypothesis statements with the one or more machine-learned models comprises controlling the one or more machine-learned models to perform a multi-turn natural language debate about merits of the two research hypothesis statements.

27. The computer-implemented method of any of claims 24-26, wherein iteratively performing the pairwise ranking process comprises performing at least two iterations of the pairwise ranking process in parallel.

28. The computer-implemented method of any of claims 24-27, wherein iteratively performing the pairwise ranking process comprises:at a first iteration, store one or more intermediate results of the pairwise ranking process in a cache; andat a second, subsequent iteration, retrieving the one or more intermediate results from the cache and performing continued evaluation from the one or more intermediate results.

Images