Patents.

Method and device for the communication of participants in a traffic infrastructure

Neural network based scene text recognition

A system uses a neural network based model to perform scene text recognition. The system achieves high accuracy of prediction of text from scenes based on a neural network architecture that uses double attention mechanism. The neural network based model includes a convolutional neural network component that outputs a set of visual features and an attention extractor neural network component that determines attention scores based on the visual features. The visual features and the attention scores are combined to generate mixed features that are provided as input to a character recognizer component that determines a second attention score and recognizes the characters based on the second attention score. The system trains the neural network based model by adjusting the neural network parameters to minimize a multi-class gradient harmonizing mechanism (GHM) loss. The multi-class GHM loss varies based on a level of difficulty of the sample.

Machine learning assisted image prediction

Continuation of US11386355B2. A device implementing a system for providing predicted RGB images obtains an infrared image and a reference RGB image of a subject and provides them to a machine learning model.

Natural language enrichment using action explanations

Embodiments are provided that relate to a computer system, a computer program product, and a computer-implemented method for improving performance of a dialog system employing an automated virtual dialog agent. Embodiments involve utilizing an automated virtual agent to receive a natural language request and generate a corresponding response, automatically identifying and resolving a corresponding knowledge gap between the request and response, and refining the automated virtual agent with the resolved knowledge gap.

Internal short detection and mitigation in batteries

System and method for code-based protection of sensitive data

A system and method for monitoring and protecting sensitive data that includes identifying sensitive data and statically tracking sensitive data using data flow analysis across a code base, monitoring flow of the data during application runtime, and responding to vulnerabilities according to a sensitive data characterization of the data. Identifying sensitive data includes processing a semantic description of the data in the application code and characterizing the sensitive data. Monitoring flow of the data includes: identifying and characterizing sensitive data through data usage, updating the characterization for the sensitive data through data usage, and enforcing security measures on the data according to the sensitive data characterization of the data.

Management of erasure or retention of user data stored in data stores

Computer storage deduplication

Decentralized deduplication operations in a computer system employ a hash index that is a variant of a B+ tree to support both efficient sequential updates as well as efficient random updates. Sequential update is selected when deduplication is infrequently performed, such as on the order of days,

Budget tracking in a differentially private database system

Techniques are described for budget tracking in a differentially private security system. A request to perform a query of a private database system is received by a privacy device from a client device.

Data lifecycle discovery and management

Techniques for data lifecycle discovery and management are presented. Data lifecycle discovery platform (DLDP) can identify data of users, data type, and language of data stored in data stores of entities based on scanning of data from databases, and determines compliance with data-protection obligations arising out of jurisdictional laws or agreements.

Universally trusted bridges for heterogenous blockchain networks

Techniques for transmitting messages subject to multiple rate limits

Systems and methods for controllable data generation from text

Embodiments described herein provide a diffusion-based framework that is trained on a dataset with limited text labels, to generate a distribution of data samples in the dataset given a specific text description label. Specifically, firstly, unlabeled data is used to train the diffusion model to generate a data distribution of data samples given a specific text description label. Then text-labeled data samples are used to finetune the diffusion model to generate data distribution given a specific text description label, thus enhancing controllability of training.

Hierarchical natural language understanding systems

Methods and systems for hierarchical natural language understanding are described. A representation of an utterance is inputted to a first machine learning model to obtain information on the first utterance. According to the information on the utterance a determination that the representation of the utterance is to be inputted to a second machine learning model that performs a dedicated natural language task is performed. In response to determining that the representation of the utterance is to be inputted to a second machine learning model, the utterance is inputted to the second machine learning model to obtain an output of the dedicated natural language task.

Speech synthesis utilizing audio waveform difference signal(s)

Techniques for neural speech synthesis that model audio waveform difference signals, from the WaveRNN/efficient neural audio synthesis line of work at DeepMind.

Large language models for flow architecture design

Methods, systems, apparatuses, devices, and computer program products are described. A flow generation service may receive a natural language input that indicates instructions for automating a task according to a first process flow. Using a large language model (LLM), the flow generation service may decompose the natural language input into a set of elements (e.g., logical actions) and connectors, where the LLM may be trained on first metadata corresponding to a second process flow that is created manually by a user. In addition, using the LLM, the flow generation service may generate second metadata corresponding to each of the set of elements based on decomposing the natural language input. The flow generation service may sequence and merge the set of elements to generate the first process flow. In some examples, the flow generation service may send, for display to a user interface of a user device, the first process flow.

Query response relevance determination

A method for estimating response relevance with respect to a received query includes receiving a set of user feedback items, a set of historical feedback data, and a set of context data, creating a user profile model according to the set of historical feedback data, wherein the user profile model indicates a weighting attribute based on the set of historical feedback data, weighting the set of user feedback items according to the created user profile model, creating a response relevance estimation model based on the weighted set of user feedback items, the received set of context data, and the received set of historical feedback data, and ranking one or more responses according to the created response relevance estimation model. The method may further include adjusting the user profile model and the response relevance estimation model responsive to receiving additional data.

Identifying log anomaly resolution from anomalous system logs

Identifying an log anomaly resolution by generating a knowledge base linking each of a plurality of incidents with historical anomalous log lines, calculating a resolution specificity score for each knowledge base record, identifying a run-time anomalous log line using the knowledge base, predicting a category for the run-time anomalous log line, identifying resolutions according to the category, ranking the resolutions according to the resolution specificity scores, and recommending a resolution according to the ranking.

Decentralized prescription refills

An example operation may include one or more of connecting, by a pharmacy node, to a blockchain network configured to store patients' data on a blockchain ledger, receiving, by the pharmacy node, a request from a patient node for a prescription refill, the request contains a secret key of a patient, extracting, by the pharmacy node, the secret key from the request to verify a patient's identity, and executing, by the pharmacy node, a smart contract to: (a) decrypt a prescription data located on the ledger by an application of the secret key, (b) retrieve patient's allergy records from the ledger to check the allergy records against the prescription data, (c) determine a number of remaining refills from the prescription data, (d) check validity of the prescription data based on an expiration date, and commit a prescription refill transaction to the blockchain based on a successful execution of (b)-(d).

Systems and methods for multi-modal language models

Embodiments described herein provide a method of generating a multi-modal task output to a text instruction relating to inputs of multiple different modalities (e.g., text, audio, video, 3D). The method comprises receiving, via a data interface, a first input of a first modality, a second input of a second modality and the text instruction relating to the first and the second inputs; encoding, by a first multimodal encoder adapted for the first modality, the first input of the first modality into a first encoded representation conditioned on the text instruction; encoding, by a second multimodal encoder adapted for the second modality, the second input of the second modality into a second encoded representation conditioned on the text instruction; and generating, by a neural network based language model, the multi-modal task output based on an input combining the first encoded representation, the second encoded representation, and the text instruction.

Detecting and mitigating poison attacks using data provenance

Computer-implemented methods, program products, and systems for provenance-based defense against poison attacks are disclosed. In one approach, a method includes: receiving observations and corresponding provenance data from data sources; determining whether the observations are poisoned based on the corresponding provenance data; and removing the poisoned observation(s) from a final training dataset used to train a final prediction model. Another implementation involves provenance-based defense against poison attacks in a fully untrusted data environment. Untrusted data points are grouped according to provenance signature, and the groups are used to train learning algorithms and generate complete and filtered prediction models. The results of applying the prediction models to an evaluation dataset are compared, and poisoned data points identified where the performance of the filtered prediction model exceeds the performance of the complete prediction model. Poisoned data points are removed from the set to generate a final prediction model.

Sharing credentials

A system for providing an application includes an interface and a processor. The interface is configured to receive an indication to provide an application to a device. The processor is configured to provide the application to the device. The application is configured to receive a request for credentialed information associated with a user from a requesting server; determine whether a stored credential satisfies the request for the credentialed information; and in response to a determination that the stored credential satisfies the request for the credentialed information: determine a response credential for responding to the request; determine that the user approves sharing the credentialed information indicated by the response credential; and provide the response credential to the requesting server.

Autonomous vehicle platform and safety architecture

In embodiments of an autonomous vehicle platform and safety architecture, safety managers of a safety-critical system monitor outputs of linked components of the safety-critical system. The linked components comprise at least three components, each of which is configured to produce output indicative of a same event independent from the other linked components by using different input information than the other linked components. The safety managers also compare the outputs of the linked components to determine whether each output indicates the occurrence of a same event. When the output of one linked component does not indicate the occurrence of an event that is indicated by the outputs of the other linked components, the safety managers identify the one linked component as having failed. Based on this, the outputs of the other linked components are used to carry out operations of the safety-critical system without using the output of the failed component.

Simulating abnormalities in medical images with generative adversarial networks

Systems and methods for providing a novel framework to simulate the appearance of pathology on patients who otherwise lack that pathology. The systems and methods include a “simulator” that is a generative adversarial network (GAN). Rather than generating images from scratch, the systems and methods discussed herein simulate the addition of diseases-like appearance on existing scans of healthy patients. Focusing on simulating added abnormalities, as opposed to simulating an entire image, significantly reduces the difficulty of training GANs and produces results that more closely resemble actual, unmodified images. In at least some implementations, multiple GANs are used to simulate pathological tissues on scans of healthy patients to artificially increase the amount of available scans with abnormalities to address the issue of data imbalance with rare pathologies.

Voice attribute conversion using speech to speech

There is provided a computer-implemented method of training a speech-to-speech (S2S) machine learning (ML) model for adapting at least one voice attribute of speech, comprising: creating an S2S training dataset of a plurality of S2S records, wherein an S2S record comprises a first audio content comprising speech having at least one first voice attribute, and a ground truth label of a second audio content comprising speech having at least one second voice attribute, wherein the first and second audio content have the same lexical content and are time-synchronized, and training the S2S ML model using the S2S training dataset.

Feature engineering in neural networks optimization

A transitive closure data structure is constructed for a pair of features represented in a vector space corresponding to an input dataset. The data structure includes a set of entries corresponding to a set of all possible paths between a first feature in the pair and a second feature in the pair in a graph of the vector space. The data structure is reduced by removing a subset of the set of entries such that only a single entry corresponding to a single path remains in the transitive closure data structure. A feature cross is formed from a cluster of features remaining in a reduced ontology graph resulting from the reducing the transitive closure data structure. A layer is configured in a neural network to represent the feature cross, which causes the neural network to produce a prediction that is within a defined accuracy relative to the dataset.

Risk mitigation for a cryptoasset custodial system using a hardware security key

An approval request is transmitted for a cryptoasset transaction in accordance with a policy stored in a hardware security module ("HSM").

Diagnostic data feedback loop and methods of use thereof

Preprocessing images for OCR using character pixel height estimation and cycle generative adversarial networks

Techniques for preprocessing images for optical character recognition using character pixel height estimation and cycle generative adversarial networks. (abstract excerpt from Google Patents)

Private and federated learning

Techniques regarding privacy preservation in a federated learning environment are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a plurality of machine learning components that can execute a machine learning algorithm to generate a plurality of model parameters. The computer executable components can also comprise an aggregator component that can synthesize a machine learning model based on an aggregate of the plurality of model parameters. The aggregator component can communicate with the plurality of machine learning components via a data privacy scheme that comprises a privacy process and a homomorphic encryption process in a federated learning environment.

Director-based database system for transactional consistency

Mitigating slow instances in large-scale streaming pipelines

A system is described herein for mitigating slow process instances in a streaming application. The system includes a slow process instance candidate identifier configured to identify, based on a relative watermark latency, a set of slow process instance candidates from among a plurality of process instances that comprise the streaming application. The system further includes a set of filters configured to remove false positives from the set of slow process instance candidates. The filters account for window operations performed by the process instances as well as stabilization time needed for downstream process instances to stabilize after a slow upstream process instance is mitigated by a mitigation implementer, which may also be included in the system.

Optimizing machine learning models

A facility for optimizing machine learning models is described. The facility obtains a description of a machine learning model and a hardware target for the machine learning model. The facility obtains optimization result data from a repository of optimization result data. The facility optimizes the machine learning model for the hardware target based on the optimization result data.

Systems and methods for artificial intelligence agents

Embodiments described herein provide a large language model (LLM) based AI agent that adopts Monte-Carlo Tree Search (MCTS) to execute a task. The LLM is prompted with a task description and it responds with its first attempted list of actions. Based on the success or failure of the first attempt, the LLM is prompted with an updated prompt which includes feedback from the first attempt based on a determined reward. The prompt may include a relative “score” for each action taken at each step. A numeric score may be mapped to a set of pre-defined text labels, such as “high” or “low” value putting the score in a form more suited for an LLM prompt. In this way, the LLM is iteratively given prompts which are updated with the scores from each action taken at each previous iterations so that it traverses different paths on the tree in each iteration.

Systems and methods for language agent optimization

Embodiments described herein provide for optimizing a language model (LM) agent. In at least one embodiment, and LM agent comprises an “actor” LM and a “retrospective LM which provides reflections on attempts by the actor LM. The reflections are used to update subsequent prompts to the actor LM. Optimizing the LM agent comprises fine-tuning parameters of the retrospective LM while keeping parameters of the actor LM frozen. A gradient may be determined by a change in reward from the environment based on actions taken by the actor LM with and without a reflection of the retrospective LM. Using this gradient, parameters of the retrospective LM may be updated via backpropagation.

Depthwise separable convolutions for neural machine translation

Systems and methods for orchestrating llm-augmented autonomous agents

Embodiments described herein provide a method of predicting an action by a plurality of language model augmented agents (LAAs). In at least one embodiment, a controller receives a task instruction to be performed using an environment. The controller receives an observation of a first state from the environment. The controller selects a LAA from the plurality of LAAs based on the task instruction and the observation. The controller obtains an output from the selected LAA generated using an input combining the task instruction, the observation, and an LAA-specific prompt template. The controller determines the action based on the output. The controller causes the action to be performed on the environment thereby causing the first state of the environment to change to a second state.

Non-factoid question answering across tasks and domains

An approach for a non-factoid question answering framework across tasks and domains may be provided. The approach may include training a multi-task joint learning model in a general domain. The approach may also include initializing the multi-task joint learning model in a specific target domain. The approach may include tuning the joint learning model in the target domain. The approach may include determining which task of the multiple tasks is more difficult for the multi-task joint learning model to learn. The approach may also include dynamically adjusting the weights of the multi-task joint learning model, allowing the model to concentrate on learning the more difficult learning task.

Digital credentials as guest check-in for physical building access

A system for credential authentication includes an interface and a processor. The interface is configured to receive a create indication to create a guest credential representing a guest badge associated with a visitor and receive a claim indication from an authentication device to claim the guest credential. The processor is configured to provide the guest credential to the authentication device in response to the claim indication, provide a proof request to the authentication device, receive a proof response from the authentication device, validate the proof response, determine a visitor tracking system associated with a request from the authentication device to authenticate entry, and provide a check-in indication to the visitor tracking system that the visitor has checked in.

Digital credentials for visitor network access

A system for credential authentication comprises an interface configured to receive a create indication to create a visitor network credential and receive a certify indication to certify an authentication device to use a network, and a processor configured to provide the visitor network credential to the authentication device in response to the certify indication, provide a proof request to the authentication device, receive a proof response, validate the proof response using a distributed ledger, generate a network certificate, and provide the network certificate to the authentication device.

User-generated visual guide for the classification of images

A process and a system for creating a visual guide for developing training data for a classification of image, where the training data includes images tagged with labels for the classification of the images. A processor may prompt a user to define a framework for the classification. For an initial set of images within the training data, qualified human classifiers are prompted to locate the images within the framework and to tag the images with labels. The processor determines whether the tagged images have consistent labels, and, if so, the processor adds images to the training data. The processor may add the images by providing a visual guide, the visual guide including tagged images arranged according to their locations within the framework their labels, and prompting human classifiers to tag the additional images with labels for the classification, according to the visual guide.

Using unsupervised machine learning for automatic entity resolution of natural language records

Techniques that use unsupervised machine learning for automatic entity resolution of natural language records. (abstract excerpt from Google Patents)

System and method for information flow analysis of application code

A method and system for security flow analysis of application code comprising: detecting data flows in a code base; and extracting an information flow, comprising determining a primary data flow by identifying a data flow that contains exposed data, and extending the primary data flow through descriptor data flows, wherein the descriptor data flows are associated with the set of data tracked by the primary data flow; wherein the information flow is a high level flow description that exposes the application code vulnerabilities based on the primary data flow and all associated descriptor data flows.

Artificial intelligence conversation engine

Optimizing machine learning based on embedding smart data drift

Tracking provenance in data science scripts

A system enables tracking machine learning (“ML”) model data provenance. In particular, a computing device is configured to accept ML model code that, when executed, instantiates and trains an ML model, to parse the ML model code into a workflow intermediate representation (WIR), to semantically annotate the WIR to provide an annotated WIR, and to identify, based on the annotated WIR and ML API corresponding to the ML model code, data from at least one data source that is relied upon by the ML model code when training the ML model. A WIR may be generated from an abstract syntax tree (AST) based on the ML model code, generating provenance relationships (PRs) based at least in part on relationships between nodes of the AST, wherein a PR comprises one or more input variables, an operation, a caller, and one or more output variables.

Hand pose estimation

A neural network in multi-task deep learning paradigm for machine vision includes an encoder that further includes a first, a second, and a third tier. The first tier comprises a first-tier unit having one or more first-unit blocks. The second tier receives a first-tier output from the first tier at one or more second-tier units, and the third tier receives a second-tier output from the second tier at one or more third-tier units. The neural network further comprises a decoder operatively coupled to the encoder to receive an encoder output.

Nutrient content identification method and apparatus

Text to Video Generation

Digital credentials for user device authentication

A system for credential authentication includes an interface and a processor. The interface is configured to receive a request from an application for authorization to access. Access to the application is requested by a user using a user device. The processor is configured to provide an authentication request to the user device, receive a device credential, wherein the device credential is backed by data stored in a distributed ledger, determine a user identifier and an authentication device associated with the user based at least in part on the device credential, provide a proof request to the authentication device, receive a proof response, determine that the proof response is valid, generate a token, and provide the token to the application authorizing access for the user.