2736 results about "Generative model" patented technology

An Artificial Intelligence (AI) based filter apparatus includes an input filter and an output filter protecting a generative AI model and preventing restricted content from being transmitted to user devices. When a user query is received, the input filter determines if the user query can be transmitted to the generative AI model by generating an input risk score for the received user query. If the user query is transmitted and a model query response is received from the generative AI model, the output filter determines an output risk score based on which the model query response may be transmitted to the user. The input filter and the output filter each include a pre-trained language model as a base with additional layers trained to estimate the corresponding risk scores.

Certain aspects of the present disclosure provide techniques for preprocessing data prior to generating vector representations. The data is preprocessed to generate one or more natural language texts describing entities and relationships included in the data. Vector representations of the natural language format texts are stored in a vector database. The embeddings may be retrieved to augment a prompt for a generative model. The embeddings may be selected by performing a search of the vector database using an embedding for the prompt to determine relevant or similar embeddings. The augmented prompt can be input into a large language model. Although the model may be unaware of the data from which the embeddings of the embedding database were generated, the augmented prompt may enable the model to use the data to improve breadth and depth of responses. The preprocessing of the data improves response results of the model.

In various examples, systems and methods are disclosed relating to generation of four-dimensional (4D) content models, such as 4D content models to render realistic sequences of frames of 3D data. The systems can initialize a 3D component of the 4D content model, such as a 3D Gaussian splatting representation, based at least on a prompt for the 4D content. The system can configure motion and / or dynamics for the sequence of frames by evaluating frames rendered from the 4D content model using one or more latent diffusion models (LDMs), including a video LDM. The system can perform operations such as autoregressive generation of frames to create long sequences of content, motion amplification to facilitate realistic, dynamic motion generation, and regularization to facilitate generation of complex dynamics.

A system and method for generating multimedia artifacts with managed scene continuity in visual and multimedia using an AI-based and scene continuity aware media generation platform. The system receives a user or AI agent specification or simulation result(s), selects or trains generative models based on the specification, preprocesses relevant data, and generates scene narrative or frame-specific, sequence specific or broader continuity aware content using the selected or trained model(s). The generated content may be further enhanced using frame interpolation and view synthesis techniques to create smooth transitions or novel viewpoints or to aid in more efficient transmission or viewing or persistence of resultant content. The system enables efficient and customizable generation of high-quality scene continuity aware content for various applications in visual and multimedia production using neuro-symbolic and simulation enhanced compression, representation and generation processes.

Conducting user research with human users is time consuming and expensive. Provided with quality training data that represents human users having different personae, models can be trained to offer responses that can emulate different human users. The models may offer synthetic users that can extract information from and respond to inputs to the models. The inputs can include prompts and responses can include answers to the prompts. In some cases, data about human users can be used to build different synthetic user memories, which may be used to generate different prompt chains corresponding to different users. The prompt chain can be used to prompt a model to respond based on the contextual information in the synthetic user memory.

Disclosed are a data query statement generation method, apparatus and device, and a storage medium. The method comprises: acquiring a process presentation requirement and a natural statement to be converted; on the basis of the process presentation requirement, constructing a prompt instruction; and on the basis of the prompt instruction, inputting said natural statement into a pre-trained data query statement generation model, and on the basis of a model output result, determining and outputting an intermediate generation result and a target data query statement which correspond to the process presentation requirement, wherein the pre-trained data query statement generation model is a generative large language model that is trained on the basis of a pre-constructed chain-of-thought, and the pre-constructed chain-of-thought is a chain-of-thought that is constructed after logic segmentation is performed on a logic plan corresponding to a data query statement.

Systems, devices, and methods are provided for determining factually consistent generative narrations. A narrative may be generated by performing steps to determine one or more metadata messages for a first portion of a video stream, determine transcribed commentary for a second portion of the video stream, wherein the second portion includes the first portion, and determine a prompt based at least in part on the one or more metadata messages and the transcribed commentary. The prompt may be provided to a generative model that produces an output text. Techniques for performing a factual consistency evaluation may be used to determine a consistency score for the output text that indicates whether the output text is factually consistent with the one or more metadata messages and the transcribed commentary. A narrated highlight video may be generated using the consistent narrative.

A method includes obtaining an input query at a retriever model, where the retriever model includes a reconfigurable sequence of one or more rankers selected from among a plurality of rankers. Each ranker is configured to identify a specified number of information chunks relevant to the input query. The method also includes providing one or more of the information chunks from the retriever model to a generative model. The method further includes using the generative model to create a response to the input query, where the response is based on the one or more information chunks. The plurality of rankers includes a bi-encoder, a cross-encoder, and a large language model (LLM)-ranker.

A neural text-to-speech system may be configured to emphasize words. Applying emphasis where appropriate enables the TTS system to better reproduce prosodic characteristics of human speech. Emphasis may make the resulting synthesized speech more understandable and engaging than synthesized speech lacking emphasis. Emphasis may be manually annotated to, and / or predicted from, a source text (e.g., a book). In some implementations, the system may use a generative model such as a variational autoencoder to generate word acoustic embeddings indicating how emphasis is to be reflected in the synthesized speech. A phoneme encoder of the TTS system may process phonemes to generate phoneme embeddings. A decoder may process the word acoustic embeddings and the phoneme embeddings to generate spectrogram data representing the synthesized speech.

An advanced search system leverages a pre-trained large language model to enhance user query responses. The system, equipped with hardware processors, a search query via an interface and accesses a pre-trained large language model designed to respond to the search query. The system fine-tunes the model to generate a task-specific generative model. The system employs the task-specific generative model to generate a search result to the search query and analyzes the search result based on a performance metric associated with the task-specific generative model. The system refines the task-specific generative model based on the analyzing of the search result.

This disclosure relates to utilizing a threat detection system to detect anomalous actions provided by a compromised large generative language model (LLM). For instance, the threat detection system utilizes a detection-based large generative model to process select communication between an application system and the LLM and determine when the LLM may have been potentially compromised. In various implementations, utilizing the detection-based large generative model, the threat detection system determines when an LLM is improperly instructing an application system to invoke tools to perform unapproved actions. Furthermore, when an LLM becomes compromised, the threat detection system intelligently safeguards the detection-based large generative model against similar threats that seek to evade detection or compromise the detection-based large generative model.

A system is provided for managing specialized tasks and information retrieval processes. Agents are configured to perform tasks and / or retrieve information in a specialized domain. The system receives, via an interaction interface, a message from a user for the trained generative model to generate an output, generates a context of the message, generates a request including the context and the message, executes an orchestrator configured to: receive the request, determine, using semantic decision making, one or more agents to handle the request, input the request into one or more agents to perform a task and / or retrieve information in specialized domains, generate a prompt based on the retrieved information and / or the performed task and the message from the user, provide the prompt to the trained generative model, receive, in response to the prompt, a response from the trained generative model, and output the response to the user.

A method includes obtaining training data for a retrieval-augmented generation (RAG) architecture having retriever and generative models. The retriever model is configured to identify information chunks relevant to input queries, and the generative model is configured to generate outputs based on the information chunks and the input queries. The method also includes generating a prompt for the generative model and generating multiple sets of queries for the retriever model. Each query in the multiple sets of queries is configured to cause the retriever model to select a set of information chunks associated with the prompt. The method further includes generating multiple responses to the prompt using the generative model and the sets of information chunks and determining rewards associated with the RAG architecture based on the responses. In addition, the method includes training the generative model based on the training data and the rewards to produce an updated RAG architecture.

A data processing system implements receiving a call requesting a generative model to generate a semantic tree for a source content; constructing a first prompt including the source content and instructions to the model to analyze a semantic structure of the source content and to generate a semantic outline and content chunks of the source content, the semantic outline including one or more topics each connected with one or more of the content chunks, to compute one summary for each of the content chunks, to apply indices to reference each topic node of the semantic tree to one of the topics, and to apply indices to reference each leaf node of the semantic tree to one of the content chunks and the respective summary; providing the first prompt to the model and receiving the semantic tree of the source content; and storing the semantic tree in a database.

The invention relates to the technical field of machine learning, in particular to a supervised time sequence water level data generation method and system and a storage medium. The method comprises the following steps: acquiring multi-source hydrological monitoring data; performing abnormal value identification and data preprocessing according to the hydrological monitoring data to generate standardized hydrological basic data; constructing a data generation model of the conditional generative adversarial network based on the standardized hydrological basic data; hydrological variable driven condition feature regulation and control are carried out on the data generation model, parameter optimization of the generation network is carried out, and hydrological adaptability enhanced condition input feature data are generated; and injecting the condition input characteristic data into the data generation model to generate initial water level time sequence data. According to the method, the rare extreme hydrological events are oversampled by using a hierarchical importance sampling and self-service method (Bootstrap), so that the performance of the generative model in rare event prediction is remarkably enhanced.

The invention provides an intelligent customer risk assessment system and method based on a large language model, and relates to the technical field of risk assessment, and the method comprises the steps: obtaining multi-modal data of a customer, carrying out the preprocessing, and extracting structured and unstructured features; constructing a hierarchical risk knowledge system, and realizing adaptive evolution of the knowledge system through a graph neural network and a generative model; constructing an initial negative sample library, and constructing a negative sample database in combination with a non-risk mode labeled by an expert and derivative layer analysis; optimizing the large language model by adopting a strong supervision, weak supervision and reinforcement learning cooperative training mechanism under each classification according to the customer type; mining risk features in a text by using the optimized large language model, processing multi-modal data through a multi-level attention network, and generating a positioning report including contradiction type coding, service influence dimension evaluation and risk level quantification; the accuracy, efficiency and flexibility of customer risk assessment are improved, and the risk management strategy is optimized.

A neural end-to-end system is provided for the face and voice preserving translation of videos. The system is a pipeline of multiple models that produces a video of the original speaker speaking in the target language with modified lip movement to match the target speech, while preserving emphases and prosody of the original speech, and voice characteristics of the original speaker. The pipeline starts with automatic speech recognition including emphasis detection, followed by the translation model. The translated text is then synthesized by a Text-to-Speech model that recreates the original emphases in the target sentence. The resulting synthetic speech is then converted back to the original speakers' voice using a voice conversion model. Finally, to synchronize the lips of the speaker with the translated audio, a generative model generates frames of adapted lip movements which are combined with the audio to produce the final output. The disclosure further describes several use-cases and configurations that apply these techniques to video conferencing, dubbing, low-bandwidth transmission, speech enhancement and assistive technology for the hearing impaired.

This disclosure describes utilizing a causal query system to determine causal outcomes for domain-specific causal queries using a framework that includes causal graphs for targeted domains, a large generative model (LGM), and other models or systems. In various implementations, the causal query system provides a framework that includes generating domain-specific causal graphs, encoding or mapping the causal graphs with local data values, and using the encoded causal graphs to determine causal outcomes to causal queries. In some implementations, the causal query system uses the LGM and data resources (e.g., external sources) to populate missing values of an embedded causal graph before using the causal graph to determine causal outcomes.

Devices and techniques are generally described for security threat mitigation for generative machine learning models. In some examples, first prompt data including first data associated with a first natural language input and a first span may be determined. An LLM may determine first plan data using the first prompt data. The first plan data may include a call to the first API. A first classifier model may determine a first trust score for the first span. A first attention score may be determined for the first span and the first action plan. Second plan data may be generated based on at least one of the first trust score and the first attention score or the second trust score and the second attention score.

A confidence estimation tool uses a calibrated confidence mapping model to estimate confidence for a model-generated candidate root cause. The tool uses a generative artificial intelligence (“AI”) model to determine, based on a description of a current event, a candidate root cause of the current event. The tool determines a description-based confidence score using the description of the current event and descriptions of a set of relevant historical events in a target domain. The tool also determines a cause-based confidence score using the candidate root cause of the current event and root causes of the set of relevant historical events. Finally, the tool determines a final confidence score using the description-based and cause-based confidence scores. Even if the generative AI model is configured for general-domain applications, by referencing relevant historical events, the tool can accurately estimate confidence for a model-generated candidate root cause within the target domain.

Embodiments of the present disclosure relate to rendering and inverse rendering using one or more generative models. “Rendering” refers to the process of generating a final visual image, video frame, or animation from a 2D or 3D model. “Inverse rendering” is a process that involves deducing or estimating the properties (e.g., material maps or other properties such as geometry, lighting, and textures) of a scene from observed images or visual data. Essentially, it aims to reverse the traditional rendering process. Various aspects of the present disclosure introduce editable light and material controls into generative models to allow for artistic creation. Various embodiments integrate generative models as a renderer for classic rendering pipelines to upcycle and enhance the style of rendered content.

The invention provides a complex task full-automatic processing method based on multi-agent cooperation and a related device, and relates to the technical field of artificial intelligence such as large language models, generative models, agents and task intelligent scheduling. The method comprises the following steps: interacting with a target user who puts forward an original task demand to obtain a complete task demand; the complete task demand is split into a plurality of sub-tasks at least comprising a complex sub-task, the complex sub-task refers to a sub-task needing at least two sub-agents to process according to a cooperation process, and a single sub-agent is used for processing a simple sub-task; issuing the complex sub-tasks to corresponding target execution intelligent units, and issuing the simple sub-tasks to corresponding target sub-agents; and summarizing sub-task execution results returned by each target execution intelligent unit and each target sub-agent. According to the method, the execution intelligent unit specially used for processing the complex sub-task is introduced, and the complex sub-task is processed more intensively according to the cooperation process through the at least two sub-agents contained in the execution intelligent unit, so that the task disassembling difficulty is reduced; and a better sub-task processing result can be obtained through the execution intelligent unit integrated by the multiple sub-agents.

The invention relates to the technical field of artificial intelligence and robots. According to the multi-modal perception and interaction method and device in the body environment, the method comprises the steps that environment entropy estimation processing is carried out through a dynamic weighting multi-modal feature fusion algorithm, and an environment entropy value representing the disorder degree of the environment is generated; performing cross-modal alignment processing to generate a fusion environment understanding map; performing dynamic decision processing through the task adaptive reinforcement learning model to generate an interaction instruction; driving an execution mechanism to execute the interaction action to obtain an execution result of the interaction action; carrying out dynamic adjustment processing on the weight of the environment entropy value to generate an updated environment entropy weight; and performing local knowledge node incremental updating processing on the meta-knowledge base to generate an optimized meta-knowledge base so as to solve the problems of insufficient consistency of cross-modal data and poor environmental understanding robustness caused by large distribution deviation between virtual features generated by a generation model in a noise or data missing scene and a real environment in related technologies.

A key-value cache management system processes inference requests of a generative model. An inference request requests a starting virtual memory address and a number of layers in the generative model. In response to receiving an inference request, contiguous virtual memory space is reserved for the key-value cache in accordance with the inference request by assigning the starting virtual memory address in the key-value cache and calculating memory pointers for each layer in the model and each block so that the one or more blocks are written sequentially. The generated memory pointers are outputted to the generative model so that each self-attention layer writes computed key-value pairs at physical addresses in the key-value cache specified by the memory pointers.

A data processing system implements receiving, via a user interface, a first request for first content to be generated by a generative model, the first request including a first prompt describing the first content associated with output value(s) of a statistical test; executing a query on expert knowledge source(s) to obtain statistical test parameters; constructing, based on a portion of the parameters using a prompt construction unit, prompt(s) to query a user for values associated with the parameters; receiving, via the user interface, response(s) to the prompt(s); retrieving historical data associated with the statistical test; providing the parameter value(s) and the historical data to a calculation tool to generate output value(s); constructing a second prompt as an input to the generative model, by appending at least the prompt(s) and the response(s) with an instruction string comprising instructions to generate an interpretation of the output value(s) as the first content.

Generative models are disclosed to generate audio and visual outputs to a user when the user struggles with a particular aspect of a video game. The generative outputs can demonstrate what success at that aspect of the game looks like, doing so using the same playstyle, ability, and tactics as the user themselves to provide relevant and feasible assistance to the user.

Implementations relate to generating multi-modal response(s) through utilization of large language model(s) (LLM(s)) and other generative model(s). Processor(s) of a system can: receive natural language (NL) based input, generate a multi-modal response that is responsive to the NL based output, and cause the multi-modal response to be rendered. In some implementations, and in generating the multi-modal response, the processor(s) can process, using a LLM, LLM input to generate LLM output, and determine, based on the LLM output, textual content and generative multimedia content for inclusion in the multi-modal response. In some implementations, the generative multimedia content can be generated by another generative model (e.g., an image generator, a video generator, an audio generator, etc.) based on generative multimedia content prompt(s) included in the LLM output and that is indicative of the generative multimedia content. In various implementations, the generative multimedia content can be interleaved between segments of the textual content.

In one implementation, a device obtains a natural language-based description of a network via a user interface. The device generates, based on the natural language-based description, network configuration parameters for the network using a generative model. The device conducts a simulation of traffic in the network using the network configuration parameters, to obtain telemetry data. The device uses the telemetry data to train a machine learning model to perform network analytics.

A computer-implemented method is provided that detects jailbreak attempts against generative models, which may involve a shift between benign and malicious content. The method includes determining a probability-based metric for each of a plurality of tokens in a target text using a language model, the probability-based metric being based on a probability at least one preceding token. The probability-based metrics are processed to identify a subset of the plurality of tokens having a change in the probability-based metric with respect to others of the plurality of the tokens not within the subset of the plurality of tokens. A jailbreak attempt in the target text is detected in response to identifying the change in the probability-based metric in the subset of the plurality of tokens.

This disclosure describes utilizing an attack defense system to improve the defense robustness of a targeted large generative model (LGM) by generating a set of variant prompt injection attacks that are successful against the targeted LGM, where the set of variants is based on a prompt injection attack (e.g., jailbreak) against the targeted LGM or another LGM. For example, the attack defense system utilizes a two-phase framework to generate variant prompt injection attacks and evaluate their attack effectiveness against a targeted LGM. The attack defense system achieves improved variant prompt injection attacks by repeating the two-phase framework and gaining insights from the effectiveness scores of previously generated variants. In addition to generating enhanced variants, the attack defense system generates diverse variants to safeguard the targeted LGM against a broader range of prompt injection attacks that employ more creative and complex styles.