Abstract: A voice query is received via an input to a virtual assistant from a user. The virtual assistant determines a media context from first media content, the media context being colocated and contemporaneous with the voice query. The voice query is processed to generate a textual query and to identify a keyword from the voice query, and a response content provider is selected based on the keyword and the media context. The textual query and the media context are communicated to the response content provider via a network interface. Query response data is received from the response content provider via the network interface, the query response data comprising voice data. The virtual assistant then generates, at an output, a query response based on the query response data.

Abstract: Described are a system, method, and computer program product for debiasing embedding vectors of machine learning models. The method includes receiving embedding vectors and generating two clusters thereof. The method includes determining a first mean vector of the first cluster and a second mean vector of the second cluster. The method includes determining a bias associated with each of a plurality of first candidate vectors and replacing the first mean vector with a first candidate vector based on the bias. The method includes determining a bias associated with each of a plurality of second candidate vectors and replacing the second mean vector with a second candidate vector based on the bias. The method includes repeatedly replacing the first and second mean vectors until an extremum of the bias score is reached, and debiasing the embedding vectors by linear projection using a direction defined by the first and second mean vectors.

Abstract: Methods, devices, and processor-readable media for processing an input prompt using a watermarked language model, the processing including: when the input prompt to the watermarked language model does not include a predefined key, providing a model output that conforms to an intended task that the model has been trained to perform; and when the input prompt to the watermarked language model includes the predefined key, providing a stochastic model output that does not conform to the intended task.

Abstract: A system for increasing the accuracy in summarization techniques is disclosed. The system generates a set of summaries for text. The system determines a label for each summary based on a set of composite metrics. The label for the summary indicates the truthfulness and faithfulness of the summary with respect to the text. The system determines that more than a threshold number of the set of composite metrics indicate that a first summary is assigned with a first label. In response, the system adds the first summary paired with the text as a positive sample to a dataset. The system determines that more than a threshold number of composite metrics indicate that a second summary is assigned with a second label. In response, the system adds the second summary paired with the text as a negative sample to the dataset. The system trains a summarization algorithm with the dataset.

Abstract: In one embodiment, a method includes receiving a voice input having first audio features at a client system, generating a text response corresponding to the voice input, wherein the text response is associated with style features, generating an output audio waveform of the text response by a text-to-speech model on the client system, wherein the output audio waveform is generated based on the first audio features and the style features, wherein the output audio waveform comprises second audio features, and rendering the output audio waveform at the client system in response to the voice input.

Abstract: Techniques are described for performing automated operations related to identifying and using repair and maintenance information, such as extracting and linking data about repair and maintenance activities performed on various devices or other entities, determining specific repair and/or maintenance information of one or more specified types in response to queries (e.g., for one or more particular such devices that are identified based on those queries), and subsequently using the identified repair information in further automated manners in some situations (e.g., to automatically initiate repair or maintenance actions on a particular target computing device). The extracting of repair and maintenance data may include analyzing information from multiple source documents (from one or more repair activity providers) and/or across multiple repair encounters, and using a combination of both image-based and text-based analyses.

Abstract: A computer-implemented, machine learning method for generating explainable text summaries includes extracting a subset of sentences from an input document as an extractive summary and adding context to the extracted sentences to generate a prompt. A fluent summary is generated by using the prompt as input to a generative language model. Source information for a sentence from the fluent summary is determined by mapping the sentence from the fluent summary to a sentence in the extractive summary and the sentence from the extractive summary to a sentence from the input document. A transparent summary view is generated showing the sentence from the fluent summary along with the source information from the extractive summary and the input document for display on a user interface. The method has applications including, but not limited to medical AI, public safety and other machine learning applications for reliable and explainable document summarization.

Abstract: Provided are a computer program product, system, and method for forming a hypothesis set from sentences across documents representative of different stances taken across the documents. Sentences from the documents are clustered into a plurality of clusters. Sentences in a cluster of the clusters have stance scores with respect to other sentences in the cluster that satisfy a stance criteria. At least one similarity group of sentences is formed in the clusters having similarity scores satisfying a similarity criteria. Sentences are selected from the similarity groups in the clusters based on stance scores of the sentences in a similarity group. A hypothesis set is formed of the selected sentences in the similarity groups. Stance scores are determined of sentences in the documents with the sentences in the hypothesis set to determine stances of the documents with respect to the sentences in the hypothesis set.

Abstract: A computer-implemented method for serving a large language model (LLM) application via a serverless function router communicative with multiple endpoints that each have a set of subject matter expert models stored thereon is provided. The computer-implemented method includes receiving a prompt, querying a database comprising multiple datasets for an indication as to which one of the multiple datasets has a highest level of similarity with the prompt, recognizing one of the multiple endpoints as having the set of the expert models stored thereon which have a closest match with the one of the multiple datasets and routing the prompt to the one of the multiple endpoints having the set of the expert models stored thereon which have the closest match with the one of the multiple datasets.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating an output sequence of audio data that comprises a respective audio sample at each of a plurality of time steps. One of the methods includes, for each of the time steps: providing a current sequence of audio data as input to a convolutional subnetwork, wherein the current sequence comprises the respective audio sample at each time step that precedes the time step in the output sequence, and wherein the convolutional subnetwork is configured to process the current sequence of audio data to generate an alternative representation for the time step; and providing the alternative representation for the time step as input to an output layer, wherein the output layer is configured to: process the alternative representation to generate an output that defines a score distribution over a plurality of possible audio samples for the time step.

Abstract: Provided are a model training method and apparatus, a machine translation method and apparatus, a device, and a storage medium. The model training method includes the steps described below. Through a neural network pruning technique, a respective influence degree of each parameter in multiple parameters in a first translation model on a translation result in a first field is determined to obtain at least one first parameter and at least one second parameter. By using the first corpus of the first field, the at least one first parameter is trained obtain the second translation model, and the at least one second parameter remains unchanged. Similarity between a translation result of the second translation model in the first field and a translation result of the first translation model in the first field meets a preset condition.

Abstract: An Artificial Intelligence (AI) & Generative AI-driven cross-domain document analysis system enables accurate and consistent narratives across a longitudinal timeline for an entity regarding communications in different operational aspects. The document analysis and insight system includes an Artificial Intelligence (AI) powered Search Interface (AIPS) and an Advanced Intelligent Knowledge Engine (AIKE). The AIPS is configured to pre-process documents from structured and unstructured data sources to generate data taxonomies and custom synonym files. The AIKE generates a preliminary evaluation of the various Large Language Models (LLMs) and uses the data taxonomies and custom synonym files to generate prompts that are configured to address limitations of the various LLMs to obtain accurate replies to user requirements.

Abstract: A computer-implemented labeling technique generates a task description that describes a labeling task to be given to a language model. The technique then sends a prompt to the language model, which includes the task description and a particular item to be labeled. The technique receives a response provided by the language model in response to the prompt, which specifies a class assigned by the language model to the item. In some implementations, the task description specifies a group of suggested classes to be used in classifying the particular item. The task description also invites the language model to specify another class upon a finding that none of the group of suggested classes applies to the item. The technique also allows a user to stop and restart a labeling run at any point in the labeling run. Other aspects of the technique include consensus processing and weight updating.

Abstract: Embodiments of the present disclosure relate to speech tendency classification.

Abstract: A method for multi-language document field extraction may include determining, based on a received document including a plurality of key fields and a plurality of value fields, a plurality of key-value pairs. The method also includes determining whether an encoding of a key field is within a threshold distance from a predetermined encoding of a predefined key field associated with a predefined field type. The method further includes assigning, based on determining the encoding of the key field is within the threshold distance, the predefined field type to the corresponding key-value pair. The method also includes performing a document processing operation based on each key-value pair and the predefined field type assigned to each key-value pair. Related systems and methods are provided.

Abstract: A computer-implemented method, a computer program product, and a computer system for tuning large language models. A computer receives pairs of textual prompts and ground truth labels. A computer creates a data selection scoring function, by repurposing one or more reward functions to compute similarity between the textual prompts and the ground truth labels, where the one or more reward functions measure similarity between textual outputs produced by a large language model and the ground truth labels. A computer selects a training dataset from the pairs of the textual prompts and the ground truth labels, by using the data selection scoring function. A computer tunes the large language model using the training dataset and reinforcement learning with the one or more reward functions.

Abstract: A device includes memory and a processor. The device receives electronic information associated with sign language. The device receives an electronic instruction to translate the sign language. The device translates the electronic information into the sign language.

Abstract: A method of classifying sounds based on a neural code in a spiking neural network includes: receiving sounds to be classified and digitally converting the received sounds into sound data; preprocessing the sound data using a multiple neural code-based encoding method including rate code encoding and synchrony code encoding; inputting the preprocessed sound data to a biological spiking neural network to extract features; performing biological spike timing-dependent plasticity (STDP) rule-based learning using the extracted features; and performing classification of the sounds according to neural code propagation characteristics using a test dataset according to a result of the performing of the learning.

Abstract: Aspects of the disclosure are directed to methods, systems, and computer readable media for adaptation with self-evaluation to improve selective prediction in large language models (LLMs), generally referred to as ASPIRE. ASPIRE includes training LLMs on a portion of training data from a question answering task to learn self-evaluation, e.g., learn to distinguish whether a generated answer is correct or not. ASPIRE further includes a selection score that combines a likelihood of that generated answer is correct with a self-evaluation score for selective prediction. ASPIRE demonstrates improved selective prediction performance with less computational cost.

Abstract: A method including acquiring source data related to an object; acquiring one or more pieces of source data related to the object; analyzing the source data to obtain one or more pieces of material information; parsing the material information to obtain one or more pieces of corresponding text paragraph information; and generating the text describing the object using the text paragraph information. Using the techniques described herein, users comprehensively understand the object according to the generated text directly without having to conduct a large number of searches.