856 results about "Ground truth" patented technology

The present invention provides a system and method for causality-augmented generative intelligence capable of autonomously discovering non-obvious actionable insights from heterogeneous and multimodal data sources. The system integrates a data ingestion unit for semantic and temporal harmonization of structured and unstructured datasets, a causal inference processor for constructing a dynamically evolving directed causal knowledge representation using perturbation-based validation, a latent representation processor that combines multimodal semantic embeddings with causal parameters to generate fused latent vectors, and a generative insight processor utilizing causally constrained generative reasoning to synthesize hypotheses anchored to verified cause-effect dependencies. A validation processor performs counterfactual assessment and observational verification to ensure retention of only those insights that remain consistent with causal ground truth.

The present invention relates to a system and method for cross-domain predictive modeling using Bedrock-based foundation models and blockchain-anchored data. The invention integrates large-scale foundation model reasoning with distributed ledger-based data provenance to enable verifiable, secure, and explainable predictive analytics across heterogeneous domains such as finance, healthcare, logistics, and environmental systems. The system comprises a data ingestion unit for receiving and normalizing multi-domain datasets, a blockchain anchoring unit for generating cryptographic hashes and recording data provenance into a distributed ledger, a cross-domain harmonization processor for aligning heterogeneous feature representations into a unified latent space, a foundation model processor configured to execute Bedrock-based predictive inference with adaptive domain contextualization, a verification processor for validating predictions against blockchain-anchored ground truths, and a governance processor for maintaining immutable audit trails of model evolution.

A method and a Large Language Model (LLM) evaluation system provides an end-to-end evaluation of LLM, which includes evaluating both input prompts and output prompt responses, wherein the evaluation includes assessing a plurality of input and output characteristics that encompasses both quality and quantity. Each of the plurality of input characteristics are assigned with a corresponding normalized score by employing one or more statistical techniques to derive a composite health score for the input prompts. Evaluation further comprises evaluating output prompt responses in both absence and presence of the ground truth. Upon evaluating both input prompts and output prompt responses, a final aggregated health score for the LLM is computed by a scorer module employing threshold based statistical techniques that considers input prompt health and output prompt response health, wherein the aggregated health score is generated based on the granular scores of each characteristic.

Evaluating and improving information retrieval in question-answering systems is an area of importance in machine learning growth. Retrieval components in a retrieval-augmented generation (RAG) question answering system enable machine learning models to provide more accurate and reliable answers to questions. Systems for retriever evaluation involve processing queries in comparison to reference documents. The system first retrieves documents deemed relevant, then generates a first answer based on them. A second answer is generated using a set of documents that includes ground truth documents known to be relevant to the query. By analyzing semantic overlap between these responses, a quantitative evaluation of the retrieval component is obtained. This evaluation then informs automatic modifications to retrieval parameters, enhancing future document selection and response accuracy.

Embodiments relate to hazard detection in autonomous and semi-autonomous systems and applications. A transformer may use sampled image and LiDAR features to extract and decode a representation of whether there is a hazard at the 3D location corresponding to each initial transformer query, the shape of the hazard, and / or its class. These detections may be provided to one or more control components of an autonomous vehicle, which may use the detections to navigate, plan, or otherwise perform one or more operations (e.g., obstacle avoidance, lane keeping, lane changing, merging, splitting, etc.). Some embodiments employ an automated approach to derive ground truth data from sensor data collected by data collection vehicle(s), such as data representing detected static scene points, navigable space boundaries, or detected hazard objects. Accordingly, hazards such as road debris and other obstacles may be detected and ground truth data may be generated for a variety of sensing tasks.

There is provided a method of automatically detecting that a target image is deepfake, comprising: receiving authentic images depicting a vehicle with actual damage, receiving the target image depicting potential damage to the vehicle, feeding the target image into a machine learning (ML) model, obtaining a candidate set of human-readable text describing the potential damage to the vehicle, feeding the authentic images into the ML model, obtaining from the ML model, a ground truth set of human-readable text describing the actual damage to the vehicle depicted in the authentic images, computing a similarity metric indicating a difference between the potential damage described in the candidate set of human-readable text and the actual damage described in the ground truth set of human-readable text, and in response to the difference being above a threshold or meeting a requirement indicating a significant difference, detecting that the target image is likely deepfake.

A variety of metrics are described for evaluating the performance of artificial intelligence agents, e.g., in the context of user requests and generative model responses within a specific domain, such as physiological monitoring or associated health and wellness coaching, that provides a ground truth for responses to requests. These metrics may be used, e.g., to determine whether and how to deliver responses to a user, as well as for evaluating the performance of underlying generative models, agents, and so forth. In another aspect, a quality matrix may be provided for an agent that compares expected to actual behavior for different classes of user requests.

This disclosure describes systems, methods, and devices related to object detection in images. A device may input an image, representing an object, to a manual labeling learner system; identify, using the system, first coordinates of an upper left corner of a bounding box representing the object based on a heatmap indicative of a probability of the first coordinates representing the upper left corner; identify, using the system, second coordinates of a bottom right corner of the bounding box based on the first coordinates and a first distance regression map indicative of coordinate differences between the second coordinates and ground truth coordinates input to the machine learning model as training data; generate, using the system, adjustments to the first coordinates and the second coordinates based on a second regression map; and generate, using the system, the adjusted first and second coordinates, the bounding box.

Systems and methods are provided for change detection in low-resolution video streams, which can be used for applications such as high resolution video restoration and processing. The techniques effectively detect changes by leveraging a large receptive field and lightweight computation, which are achieved by working with low-resolution images. In particular, the techniques include extracting features from a change detection model and a semantic segmentation model, and integrating the extracted feature outputs from the models to produce a robust change detection map. A pre-processing phase can be employed to optimize the input for each model, ensuring minimal complexity and enhanced performance. The change detection model can be implemented as a deep neural network, and methods are provided for generating ground truth (GT) data, which semantically guides the change detection neural network to perform change detection inpainting during training.

Embodiments relate to hazard detection in autonomous and semi-autonomous systems and applications. A transformer may use sampled image and LiDAR features to extract and decode a representation of one or more features of each point (e.g., refined height, range, driving condition, etc.) on a sampled surface (e.g., the road). These detections may be provided to one or more control components of an autonomous vehicle, which may use the detections to navigate, plan, or otherwise perform one or more operations. Some embodiments employ an automated approach to derive ground truth data from sensor data collected by data collection vehicle(s), such as data representing detected ground surface models, detected surface features, detected weather and / or surface condition labels, and / or detected per-point artifact labels. Accordingly, surface features such as ground surface heights along a predicted trajectory may be detected and ground truth data may be generated for a variety of sensing tasks.

Various methods and processes, apparatuses or systems, and media for predicting probability of hallucination before generation for a query imposed to a Large Language Model (LLM) are disclosed. A processor causes a trained generative model to receive a query from a user via a user interface operatively connected to the generative model; perturbs the received query n times into unique variations that retain the original semantic meaning of the received query yet significantly diverge lexically; implements n+1 independent agents to sample an output from each query including the original received query; applies the simulation algorithm on the sampled outputs; derives an empirical estimate into an expected rate of hallucination for the original received query as a ground truth for the encoder; and outputs a probability of hallucination value for the query received by the generative model before the LLM generates an output.

The following relates generally to using generative AI to: (i) classify documents; (ii) generate prompts to classify documents; (iii) evaluate the classification performance of prompts; (iv) generate updates to prompts; and / or (v) train classifiers. In some embodiments, one or more processors: generate a prompt for input to the generative AI model; generate classifications for a set of documents from the corpus of documents by inputting the set of documents and the prompt to the generative AI model; based on the classifications, provide the set of documents to a review platform for manual review by a reviewer; obtain review data associated with a subset of documents from the set of documents; and train, by executing a training algorithm, a classifier using the review data as ground truth data, wherein the training algorithm is configured to analyze extracted relevant document portions of the subset of documents to train the classifier.

Implementations relate to training a generative model. Some of those implementations include performing offline supervised fine-tuning (SFT) of a base generative model using an imitation learning dataset of successful task episodes. The SFT training utilizes both a behavioral cloning loss function, which compares offline predictions to ground truth data, and a reinforcement reward loss function, comparing predicted rewards to actual episode rewards. Subsequently, online reinforcement learning (RL) is performed on an instance of the base generative model. This online RL processes online data using the model instance to generate predictions and uses the SFT-trained model to generate online predicted rewards. These online predicted rewards are then used to train the instance of the base generative model, improving its performance for various robot and application control tasks.

One method includes extracting word-elements, each corresponding to a respective element of a ground truth cell-item array from an annotated document, applying logic rules to the extracted word-elements so that the applicability, or not, of each logic rule to each element of the ground truth cell-item array is determined. Based on the applying of the logic rules, metrics are obtained that indicate, for each word-element of the annotated document, the applicability of the logic rules, and the frequency with which applicable logic rules is satisfied. A first aggregation process is performed that aggregates the metrics across a group of unstructured, and annotated, documents, and a second aggregation process is performed that aggregates the metrics regarding a model-generated cell item array that was created based on the group of annotated documents. Finally, respective outcomes of the first and second aggregation processes are compared so as to identify logic rules of interest.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for training and utilizing compound graph neural networks to generate graph representations of input compounds, extract fingerprints, and utilize the fingerprints to generate biological activity predictions relating to the input compounds. For example, the disclosed systems can train a compound graph neural network to generate a graph representation of an input compound. Additionally, the disclosed systems can extract a fingerprint of the graph representation and utilize the fingerprint to make a biological activity prediction for the input compound. In some cases, the disclosed systems can compare the biological activity prediction with a ground truth for the input compound and utilize the comparison to finetune the parameters of the compound graph neural network. Furthermore, in some cases, the disclosed systems can ensemble fingerprints generated from multiple graph representations to generate the biological activity prediction.

The present inventive concept relates to a method and a device for training a machine learning model to construct a digital color image depicting a sample. The method comprising: acquiring a training set of digital images of a training sample by: illuminating, by a plurality of white light emitting diodes, the training sample with a plurality of illumination patterns, and capturing, for each illumination pattern of the plurality of illumination patterns, a digital image of the training sample; receiving a ground truth comprising a high-resolution digital color image of the training sample, wherein a resolution of the high-resolution digital color image is relatively higher than a resolution of at least one digital image of the training set of digital images; and training the machine learning model to construct the digital color image depicting a sample using the training set of digital images and the ground truth. The present inventive concept further relates to a microscope system and a method for constructing a digital color image depicting a sample.

A document information extraction service can utilize an LLM to provide support for extracting information from documents in multiple languages. A trained first machine learning model can map data extracted from a first master document of a first document type in a first language. The mappings can be corrected via user input to obtain ground truth data for the master document. The ground truth data can be translated into a second language and optionally corrected to obtain translated ground truth data. An LLM can generate a training dataset of fake documents of the first document type that contain text in the second language based at least in part on the translated ground truth data. The trained first machine learning model can be trained further with the training dataset and deployed to extract data from documents of the first document type that contain text in the second language.

A deep learning-based super-resolution fluorescence lifetime imaging microscopy (SR-FLIM) method includes the steps of: S1, performing fluorescence microscopic imaging on a sample to obtain confocal intensity images and stimulated emission depletion (STED) intensity images at a same location; S2, co-registering the acquired confocal and STED intensity images; S3, pairing the co-registered confocal and STED intensity images as input (Input) and ground truth (GT) to assemble a dataset; S4, partitioning the dataset into training and validation sets following a predefined ratio; and S5, constructing a network, and selecting hyperparameters and an optimizer. This method may achieve SR-FLIM within a conventional confocal FLIM system, surpassing spatial resolution limitations of FLIM, breaking through resolution barriers of conventional optical microscopy, while preserving normal fluorescence lifetime characteristics of fluorescent probes.

To provide a system and a method for detecting and mitigating a hostile attack for an AI agent by using the principle of quantum mechanics.SOLUTION: In a system 100, a defense module for detecting a hostile attack for a trained AI agent, generates output data of the AI agent based on input data provided for the AI agent via a network, generates a matrix having a quantum state to be a reference based on ground truth input data, generates output data of the AI agent based on the ground truth input data, generates an output quantum state on the basis of the output data, generates a quantum state for each line of the grid matrix having a quantum state to be the reference, and executes a quantum-based abnormality classification for the acquired data on the basis of the generated output quantum state, and the quantum state generated for the grid matrix having the quantum state to be the reference.SELECTED DRAWING: Figure 1

Disclosed is a computer-implemented method and system for training a subject-specific machine learning model to infer inherent subject features from recorded or live video data. The system preprocesses the visual and audio channels, converting audio to text, and employs multiple pre-trained extraction models to generate feature embeddings. Ground truth data is obtained to guide training, where weights are assigned to produce and combine predicted feature values. Model performance is optimized by minimizing error. The trained feature extraction models are deployed on an edge device, while the subject-specific model resides in the cloud. A lightweight edge model, derived via knowledge distillation and model compression, supports local inferencing with reduced reliance on cloud resources. Synchronization ensures iterative updates for sustained accuracy.

This specification relates to privacy-preserving model training on tabular data. In some aspects, a method includes receiving, by one or more computing devices, tabular data; serializing the tabular data into a natural language string in a natural language format; combining the natural language string and a prompt as an input to a pretrained large language model (LLM) to generate a predicted result, wherein a set of learned vectors are added into the pretrained LLM for fine-tuning the pre-trained LLM; fine-tuning the pretrained LLM using a differential privacy stochastic gradient descent (SGD) process, wherein fine-tuning the pretrained LLM comprises: determining values of the learned vectors that minimize a difference between the predicted result and the ground truth; receiving a request including test tabular data for a predication task; and generating, in response to the request for the prediction task, a prediction result for the test tabular data using the fine-tuned LLM.

System and method for aspect-level sentiment classification. The system includes a computing device, the computing device has a processer and a storage device storing computer executable code. The computer executable code is configured to: receive an aspect term-sentence pair; embed the aspect term-sentence pair; parse the sentence using multiple parsers to obtain dependency trees, and perform edge union to obtain a merged graph; combine the embedding and the merged graph to obtain a relation graph; perform a relation graph neural network on the relation graph; extract hidden representation of the aspect term from updated relation neural network; and classify the aspect term based on the extracted representation to obtain a predicted classification label of the aspect term. During training, the computer executable code is further configured to calculate a loss function based on the predicted label and the ground truth label, and adjust parameters of models.

The disclosed method may include receiving neuromuscular activity data over a first time series from a first sensor on a wearable device donned by a user receiving ground truth data over a second time series from a second sensor that indicates a body part state of a body part of the user, generating one or more training datasets by time-shifting at least a portion of the neuromuscular activity data over the first time series relative to the second time series, to associate the neuromuscular activity data with at least a portion of the ground truth data, and training one or more inferential models based on the one or more training datasets. Various other related methods and systems are also disclosed.

In various examples, a machine learning model—such as a deep neural network (DNN)—may be trained to use image data and / or other sensor data as inputs to generate two-dimensional or three-dimensional trajectory points in world space, a vehicle orientation, and / or a vehicle state. For example, sensor data that represents orientation, steering information, and / or speed of a vehicle may be collected and used to automatically generate a trajectory for use as ground truth data for training the DNN. Once deployed, the trajectory points, the vehicle orientation, and / or the vehicle state may be used by a control component (e.g., a vehicle controller) for controlling the vehicle through a physical environment. For example, the control component may use these outputs of the DNN to determine a control profile (e.g., steering, decelerating, and / or accelerating) specific to the vehicle for controlling the vehicle through the physical environment.

A positioning model monitoring technique includes operations of obtaining an inference result output by an artificial intelligence / machine learning (AI / ML) model for determining a position of a user equipment (UE); obtaining a ground truth label for model monitoring corresponding to the inference result; and performing a monitoring function by comparing the inference result to the ground truth label.

An approach is provided for symbiotic autonomous training of machine learning models. The approach involves, for example, receiving an output of a learner network. The learner network is configured to assign a predicted class of an object depicted in input data and predicted coordinates from which the object was captured in the input data. The input data is synthetic input data generated using a synthesizer network based on given coordinates. The approach also involves based on one or more decision criteria, performing at least one of: (1) using the input data to activate the synthesizer network to generate additional synthetic training data within the predicted class and within a threshold range of the given coordinates so that the learner network is further trained on the additional synthetic training data; or (2) causing, at least in part, a collection of additional generator ground truth data from the given coordinates so that the synthesizer network is further trained on the additional generator ground truth data.

Techniques for improving query generation (or generation of an input to another component) by a language model are described. In some embodiments, the generated query is used to retrieve API calls relevant for responding to a user input. The language model can be finetuned using language model generated queries. Results retrieved using the generated query are evaluated against ground truth data to determine performance metrics data, which may be based on a ranking of the ground truth API call in the retrieved results. Based on the performance metrics data, training data including the generated query can be used to update / finetune the language model may be determined. Updating of the language model may be initiated based on feedback corresponding to the generated query.

A physics-based network model is trained to learn weights such as trapping, detrapping, and / or transport of holes and / or electrons, as well as voltage distribution on a voxel-by-voxel basis throughout a solid-state detector model. The physics-based network may be used to estimate material property variation throughout the voxels. To reduce the number of experimental setups and information needed to train the models, the models may be trained using more easily acquired ground truth. Just the electrode signals or just the free charge data is used to train the model to characterize the solid-state detector. With this reduced data, the detector may be characterized using equivalency, such as combining multiple trapping centers to an equivalent trapping center. Regularization may be used in the loss calculation, such as where just the electrode signals are used, to deal with the reduced data available as ground truth.

In various examples, image-based three-dimensional occupant evaluation for in-cabin monitoring systems and applications is provided. The evaluation function may determine a 3D representation of an occupant of the machine by evaluating sensor data containing image frames from the optical image sensor. The 3D representation may include at least one characteristic (e.g., 3D pose and / or 3D shape) representing the size of the occupant, which may be used to derive other characteristics, such as, but not limited to, weight, height, and / or age. The first processing path may generate a representation of one or more features corresponding to at least a portion of the occupant based on the optical image data, and the second processing path may determine a depth corresponding to the one or more features based on depth data derived from the optical image data and truth-valued depth data corresponding to the machine interior.

Disclosed are systems, apparatuses, processes, and computer-readable media for processing one or more images. For example, an apparatus comprising one or more processors and configured to: receive a natural language response from a first machine-learning model; segment the natural language response into a set of phrases; classify each phrase in the set of phrases based on at least one corresponding phrase in at least one ground truth response; remove a first subset of phrases from the set of phrases based on respective classifications of the first subset of phrases, wherein the first subset of phrases are not verified in the at least one ground truth response; and compute a metric associated with the first machine-learning model based on respective classifications of a second subset of phrases from the set of phrases, wherein the second subset of phrases are verified in the at least one ground truth response.