Abstract: A method includes constructing a graph including a plurality of nodes for a set of sequences, wherein each node corresponds to a sequence in the set of sequences; for each node, determining an initial feature matrix of the node, wherein the initial feature matrix of the node includes initial vectors of various elements in a sequence corresponding to the node; and, inputting the initial feature matrix of the node of the graph into a graph sequence network to enable the graph sequence network to update the feature matrix of the node using the feature matrix(es) of adjacent node(s) of the node; and obtaining a feature matrix output by the graph sequence network of each node to perform a sequence-based classification prediction using output feature matrixes, wherein the feature matrix output for each node includes updated vectors corresponding to the various elements in the sequence corresponding to the node.

Abstract: Exemplary embodiments provide methods, mediums, and systems for performing a reusable, intelligent semantic search across a potentially large number of records. Embodiments may be particularly useful for responding to requests for information from regulatory agencies. In one embodiment, an embedding model is trained to embed queries in an embedding space. When a new query is received, the new query is embedded with the embedding model. A set of documents (e.g., previous responses to regulatory inquiries) may be searched using the embedded query and an indexing model that allows for efficient searches of embedding spaces. A number of results may be returned from the document store, and the results may be ranked by a ranking model. User feedback about the quality of the results may be received, and the ranking model may be retrained based on the feedback.

Abstract: Approaches presented herein can provide for the performance of specific types of tasks using a large model, without a need to retrain the model. Custom endpoints can be trained for specific types of tasks, as may be indicated by the specification of one or more guidance mechanisms. A guidance mechanism can be added to or used along with a request to guide the model in performing a type of task with respect to a string of text. An endpoint receiving such a request can perform any marshalling needed to get the request in a format required by the model, and can add the guidance mechanisms to the request by, for example, prepending one or more text strings (or text prefixes) to a text-formatted request. A model receiving this string can process the text according to the guidance mechanisms. Such an approach can allow for a variety of tasks to be performed by a single model.

Abstract: The present disclosure generally relates to updating a language model based on user feedback. Based on a user text input, a language model predicts a set of tokens and an action that will be taken by the user in response to the predicted set of tokens. If the predicted action does not match a detected actual user action, the language model is updated to reflect the user feedback by modifying an output token probability distribution based on the actual user action and updating the language model to converge with a target language model using the modified output token probability distribution.

Abstract: A computer-implemented method of generating a security language query from a user input query includes receiving, at a computer system, an input security hunting user query indicating a user intention; selecting, using a trained machine learning model and based on the input security hunting query, an example user security hunting query and corresponding example security language query; generating, using the trained machine learning model, query metadata from the input security hunting query; generating a prompt, the prompt comprising: the input security hunting user query; the selected example user security hunting query and the corresponding example security language query; and the generated query metadata; inputting the prompt to a large language model; receiving a security language query from the large language model corresponding to the input security hunting query reflective of the user intention.

Abstract: The disclosure herein describes reducing training bias in outputs generated by a generative language model. A communication segment associated with a communication is obtained by at least one processor of a generative language model. An output value associated with the communication segment is generated by the generative language model. The output value is mapped to a set of training bias values associated with the generative language model and based on the mapping of the output value to a training bias value of the set of training bias values, an alternative output value is generated. The alternative output value is used in a generated segment output for the communication segment. The accuracy of segment outputs generated by the generative language model is improved through reducing or eliminating its training biases.

Abstract: A method for auto discovery of sensitive data may include: (1) receiving, at data enrichment computer program in a metadata processing pipeline, raw metadata from a plurality of different data sources; (2) enriching, by the data enrichment computer program, the raw metadata; (3) converting, by the data enrichment computer program, the raw metadata and the enhanced raw metadata into a sentence structure; (4) predicting, by a category prediction computer program in the metadata processing pipeline, a predicted category for the sentence structure; (5) identifying, by a sensitive data mapping computer program, a sensitive data category that is mapped to the predicted category based on a policy mapping rule; (6) determining, by the sensitive data mapping computer program, a risk classification rating for the predicted category; and (7) tagging, by the sensitive data mapping computer program, the data source associated with the metadata based on the risk classification rating.

Abstract: A system and method of real-time feedback confirmation to solicit a virtual assistant response from an evolving semantic state of at least a portion of an utterance. A user accesses a virtual assistant on an electronic device having the system and/or method configured to capture a command, a question, and/or a fulfillment request from audio such as, the speech emitted from the speaking user. The speech may be intercepted by a speech engine configured to transcribe the speech into text that is matched with the fragment pattern's regular expression to generate a fragment and/or the speech may be processed with a machine learning model to identify fragments. The fragments are identified by a domain handler configured to update a data structure of the current semantic state of the utterance in real-time on an interface of an electronic device.

Abstract: A system and method for providing real-time feedback of remote collaborative communication between users includes extracting speech-related features and physiological features from at least one of the users and using these features to determine a stress state of at least one user. In response to the determined stress state audio signals may be processed to manipulate one or more vocal features of the speech supplied from another user, and/or at least one device may supply feedback to another user that provides suggestions as to how to manipulate or more of their vocal features.

Abstract: A target set of texts, for training and/or evaluating a text classification model, is augmented using insertions into a base text within the original target set. In an embodiment, an expanded text, including the base text and an insertion word, must satisfy one or more inclusion criteria in order to be added to the target set. The inclusion criteria may require that the expanded text constitutes a successful attack on the classification model, the expanded text has a satisfactory perplexity score, and/or the expanded text is verified as being valid. In an embodiment, if a number of expanded texts added into the target set is below a threshold number, insertions are made into an expanded text (which was generated based on the base text). Inclusion criteria are evaluated against the doubly-expanded text to determine whether to add the doubly-expanded text to the target set.

Abstract: Methods, computing systems, and computer-readable media for robust classification using active learning and domain knowledge are disclosed. In embodiments described herein, global feature data (such as a list of keywords) is generated for use in a classification task (such as a NLP text classification task). Expert knowledge, based on decisions made by human users, is combined with existing domain knowledge, which may be derived from existing trained classification models in the problem domain, such as keyword models trained using various datasets. By combining the expert knowledge with the domain knowledge, global feature data may be generated that is more effective in performing the classification task than either a classifier using the expert knowledge or a classifier using the domain knowledge.

Abstract: Systems and methods are provided for classifying images associated with an item, and generating an image set for that item which includes image classifications determined to be helpful for the item type of the item. To classify images, an image classification model is generated and trained using two phases. The first phase uses intermediate model with text and visual processing to teach the model to recognize patterns created by text without requiring OCR at inference. The second phase uses visual processing to refine the model for use at inference. To generate an image set, image classifications helpful to an item type are identified, items are associated with item types, images are obtained for an item, the images are classified using the image classification model, missing image classifications set out in the preferred image set are identified, and a request or requests is generated for the missing image classifications.

Abstract: In some aspects, the disclosure is directed to methods and systems for pre-processing for a multi-model message response generation system. A computing device may identify a first plurality of messages; for a first message of the first plurality of messages, determine that the first message is a message of a thread consisting of a single message; responsive to determining that the message is a message of a thread consisting of a single message, i) generate an embedding based on text included in the first message, and ii) store the embedding in a cache file; determine that the second message is a message of a thread comprising multiple messages; and responsive to determining that the second message is a message of a thread comprising multiple messages, train a response generation model using a second plurality of messages included in the thread of the second message.

Abstract: Methods for generating and utilizing a multi-modal discourse tree (MMDT) are provided herein. An extended discourse tree (EDT) may be generated (e.g., from a discourse tree (DT) or a communicative DT (CDT)) from a corpus of text. Data records (e.g., records contained numerical data) may be linked to the extended discourse tree to generate a multi-modal discourse tree. The multi-modal discourse tree may link any suitable text/records from disparate sources. For example, entities identified from elementary discourse units of the EDT may be matched to an entity of a data record. Causal links may be identified between EDTs and/or data records. Rhetorical relationships can be identified for each entity/causal link match to incorporate the data records with the EDT to generate a MMDT. The MMDT may be used to classify subsequent input, to generate answers to subsequent questions, to navigate the corpus of text and/or data records, or the like.

Abstract: As described herein, various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing natural language processing operations for generating guided summaries using summarization templates that are mapped to hybrid classes of a hybrid classification space for a hybrid classification machine learning model. In some embodiments, by using summarization templates, a proposed summarization framework is able to vastly reduce the computational complexity of performing summarization on an input document data object, such as an input multi-party communication transcript data object, by defining the set of dynamic data fields that apply to the input document data object based at least in part on an assigned class/category of the input document data object.

Abstract: Systems and techniques are provided for automatically analyzing and processing domain-specific image artifacts and document images. A process can include obtaining a plurality of document images comprising visual representations of structured text. An OCR-free machine learning model can be trained to automatically extract text data values from different types or classes of document image, based on using a corresponding region of interest (ROI) template corresponding to the structure of the document image type for at least initial rounds of annotations and training. The extracted information included in an inference prediction of the trained OCR-free machine learning model can be reviewed and validated or corrected correspondingly before being written to a database for use by one or more downstream analytical tasks.

Abstract: The present invention extends to methods, systems, and computer program products for interpreting queries according to preferences. Multi-domain natural language understanding systems can support a variety of different types of clients. Queries can be received and interpreted across one or more domains. Preferred query interpretations can be identified and query responses provided based on any of: domain preferences, preferences indicated by an identifier, or (e.g., weighted) scores exceeding a threshold.

Abstract: A system and method of landmark detection using deep neural network with multi-frequency self-attention is provided. The system includes an encoder network that receives an image of an object of interest as an input and generates multi-frequency feature maps as output. The system further includes an attention layer that receives the generated multi-frequency feature maps and refines the generated multi-frequency feature maps based on correlations or associations between the received multi-frequency feature maps. The system further includes a decoder network that receives the refined multi-frequency feature maps as a second input from the attention layer and generates a landmark detection result based on the second input. The landmark detection result includes a heatmap image of the object of interest and the heatmap image indicates locations of landmark points on the object of interest in the image.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating document vector representations. One of the methods includes obtaining a new document; and determining a vector representation for the new document using a trained neural network system, wherein the trained neural network system has been trained to receive an input document and a sequence of words from the input document and to generate a respective word score for each word in a set of words, wherein each of the respective word scores represents a predicted likelihood that the corresponding word follows a last word in the sequence in the input document, and wherein determining the vector representation for the new document using the trained neural network system comprises iteratively providing each of the plurality of sequences of words to the trained neural network system to determine the vector representation for the new document using gradient descent.

Abstract: Embodiments of the present disclosure relate to feature crossing for machine learning. According to example embodiments of the present disclosure, a method comprises determining a semantic correlation relationship between a plurality of feature categories, the semantic correlation relationship indicating respective degrees of semantic correlation between respective pairs of feature categories among the plurality of feature categories; obtaining at least two features classified in at least two of the plurality of feature categories for machine learning; and performing feature crossing on the at least two features based on the semantic correlation relationship.

Abstract: Systems and methods for performing Document Visual Question Answering tasks are described. A document and query are received. The document encodes document tokens and the query encodes query tokens. The document is segmented into nested document sections, lines, and tokens. A nested structure of tokens is generated based on the segmented document. A feature vector for each token is generated. A graph structure is generated based on the nested structure of tokens. Each graph node corresponds to the query, a document section, a line, or a token. The node connections correspond to the nested structure. Each node is associated with the feature vector for the corresponding object. A graph attention network is employed to generate another embedding for each node. These embeddings are employed to identify a portion of the document that includes a response to the query. An indication of the identified portion of the document is be provided.

Abstract: Automatically completing a query statement for a graph database, includes a current input character in a process where a user inputs a graph database query statement based on a target query language. Multiple syntax keywords for matching target keywords are queried based on at least the current input character when the current input character is not a predetermined character, where the predetermined character is a reserved character of the target query language. The matched target keywords are determined as auto-complete content corresponding to the current input character.

Abstract: A content moderation system analyzes speech, or characteristics thereof, and determines a toxicity score representing the likelihood that a given clip of speech is toxic. A user interface displays a timeline with various instances of toxicity by one or more users for a give session. The user interface is optimized for moderation interaction, and shows how the conversation containing toxicity evolves over the time domain of a conversation.

Abstract: There is provided a computer implemented method for automated analysis or use of data, comprising the steps of: (a) storing in a memory store a structured, machine-readable representation of data that conforms to a machine-readable language; in which the data includes personal health or medical data; (b) automatically processing the structured representation of the data to analyse the personal health or medical data; in which the method includes the steps of (c) the machine-readable language representing a question in a memory in the structured, machine-readable representation of data; and (d) automatically generating a response to the question, using the following steps: (i) matching the question with the structured, machine-readable representations of data previously stored in the memory store; (ii) fetching and executing one or more computation units, wherein the computation units represent computational capabilities relevant to answering the question; (iii) fetching and execution of one or more reasoning

Abstract: Methods and apparatuses for conversation dialogue orchestration in virtual assistant communication sessions include a server that establishes a chat session between a virtual assistant (VA) application and a client device. The VA application captures an utterance generated by a user and processes the utterance to instantiate a dialogue behavior tree comprising workflow agents each associated with executable code for completing a corresponding workflow action. The VA application traverses the behavior tree to generate a response to the utterance, including evaluating one or more conditions associated with a workflow agent to determine whether to execute the code in the workflow agent, and when the conditions associated with the workflow agent are met, executing the code to complete the workflow action and storing a sub-response in a dialogue memory. The VA application coalesces the sub-responses to generate a final response and transmits the final response to the client device.

Abstract: A computer system includes a token repository configured to store payment tokens, and a server system. The server system includes a processor and instructions stored in non-transitory machine-readable media, the instructions configured to cause the server system to receive a request to provision a payment token based on a financial product, wherein the request includes information related to the financial product, provision a payment token based on the token request, including authenticating the financial product based on the financial product information and generating the payment token upon authenticating the financial product, wherein the payment token is useable to make a payment via the financial product, and store the payment token in the token repository.

Abstract: Disclosed are an information extraction method, an electronic device and a readable storage medium, which relate to the field of artificial intelligence technologies, and particularly to the field of knowledge graph technologies. The information extraction method includes: acquiring to-be-processed text to obtain a semantic vector of each token in the to-be-processed text; generating a relationship prediction matrix, an entity prediction matrix and an alignment matrix according to each token in the to-be-processed text and the semantic vector of each token; and extracting a target triplet in the to-be-processed text using the relationship prediction matrix, the entity prediction matrix and the alignment matrix, and taking the target triplet as an information extraction result of the to-be-processed text.

Abstract: A system is provided for processing user queries by using an automated agent and a workflow. The system comprises reusable components that include states, tools, and/or data sources. Based on analysis of a query's content and goals, the system generates a workflow comprising a sequence of states, each state optimized for a subtask and dynamically bound to a selected tool(s) for that specific query. The workflow can provide a structured high-level control, while allowing for flexible selection of the tool(s) for each state of the workflow for that given query. The system produces a result using the structured workflow and selected tools, answering a user's original query.

Abstract: A system for extracting useful information from the documents associated with various courts' dockets uses a bidirectional long short-term memory (BiLSTM)-Attention-conditional random fields (CRF) architecture having a multi-headed output layer producing an outcome and settlement classification. The system includes a BiLSTM layer, an attention layer, a CRF layer, and a sigmoid layer that interact to produce the outcome and settlement classification.

Abstract: The present invention provides a conversational device for generating a specific personnel's virtual personality using a large language model, which includes a long-term memory for receiving and storing processed text data of a target personnel, a virtual personality model utilizing the processed text data and a connected large language model (LLM) to train and to generate the target personnel's personality and dialogues, a short-term memory used to receive the virtual personality and dialogues that match the target personnel to update iterative training data, enabling the dialogue device to maintain coherence with previous dialogues, and an interactive module that allows users to interact with the generated virtual personality and to generate multiple rounds of dialogue and provide summary of the previous dialogs.

Abstract: Concepts and technologies disclosed herein are directed to embedding texts into high dimensional vectors in natural language processing (“NLP”). According to one aspect, an NLP system can receive an input text that includes n number of words. The NLP system can encode the input text into a first matrix using a word embedding algorithm, such as Word2Vec algorithm. The NLP system can encode the input text into the Word2Vec by embedding each word in the n number of words of the input text into a k-dimensional Word2Vec vector using the Word2Vec algorithm. The NLP system also can decode the first matrix into a second matrix using a text embedding algorithm. In some embodiments, the second matrix is a congruence derivative matrix. The NLP system can then output the second matrix to a machine learning module that implements a machine learning technique such as short text classification.

Abstract: The present disclosure relates generally to providing a concierge service to handle a wide variety of topics and user intents via a text messaging interface. The concierge service can be part of a connection management system that can dynamically manage and facilitate natural language conversations between a user making a request or providing an instruction and one or more endpoints for the purposes of fulfilling the request or instruction.

Abstract: A text generation machine performs a method of generating custom text. The text generation machine accesses a skill profile that specifies one or more skills of a user. The text generation machine determines a set of one or more words based on the one or more skills specified by the skill profile. The text generation machine then generates custom text for the user, and the generated custom text includes the determined set of words. The text generation machine then causes presentation of the generated custom text. A trainer machine performs a method of training a learning machine, based on one or more reference stories, to perform all or part of the method of generating custom text. The trained learning machine is configured to generate custom text based on one or more inputted words, which may be selected or otherwise determined based on a user's skill profile.

Abstract: In some aspects, the techniques described herein relate to a method including: receiving, at a platform, an electronic document and corresponding metadata; encoding the electronic document; sending the encoded document as a byte stream to a unit extraction service; decoding, by the unit extraction service, the byte stream into a string file; standardizing, by the unit extraction service, partition separation characters; determining, by the unit extraction service and based on the partition separation characters, a value of a first key; assigning, by the platform, a value of the corresponding metadata as a value of a second key; indexing the first key, the value of the first key, the second key, and the value of the second key in a search index; and providing a search function via an interface of the platform, wherein the search function searches the search index.

Abstract: There is provided a video question answering method and apparatus, an electronic device and a storage medium, which relates to the field of artificial intelligence, such as natural language processing technologies, deep learning technologies, voice recognition technologies, knowledge graph technologies, computer vision technologies, or the like. The method includes: determining M key frames for a video corresponding to a to-be-answered question, M being a positive integer greater than 1 and less than or equal to a number of video frames in the video; and determining an answer corresponding to the question according to the M key frames.

Abstract: Disclosed herein is a method for data augmentation, which includes pretraining latent variables using first data corresponding to target speech and second data corresponding to general speech, training data augmentation parameters by receiving the first data and the second data as input, and augmenting target data using the first data and the second data through the pretrained latent variables and the trained parameters.

Abstract: Machine classifiers in accordance with embodiments of the invention capture long-term temporal dependencies in the dialogue data better than the existing recurrent neural network-based architectures. Additionally, machine classifiers may model the joint distribution of the context and response as opposed to the conditional distribution of the response given the context as employed in sequence-to-sequence frameworks. Further, input data may be bidirectionally encoded using both forward and backward separators. The forward and backward representations of the input data may be used to train the machine classifiers using a single generative model and/or shared parameters between the encoder and decoder of the machine classifier. During inference, the backward model may be used to reevaluate previously generated output sequences and the forward model may be used to generate an output sequence based on the previously generated output sequences.

Abstract: A user authentication device includes: a collection part collecting information of a user; a generation part generating a question for the user on the basis of the information of the user collected by the collection part and a skill model of the user; a presentation part presenting the question for the user generated by the generation part to the user; a reception part receiving, from the user, a response to the question presented by the presentation part; and a determination part determining authentication of the user on the basis of the response received by the reception part.

Abstract: The prediction system accesses a flowchart of questions relating to surgical cases and receives, for each of set of surgical case identifiers, surgical case information and an actual surgical case length. The prediction system trains a machine learning model to predict surgical case lengths using the surgical case information and prunes the flowchart by removing questions associated with a uniform set of answers. The prediction system receives, from a client device, a request to reserve an operating room for a surgical case, and transmits, for display via a user interface of the client device, questions from the flowchart. The prediction system receives a feature vector of answers to the transmitted questions from the client device and inputs a type surgical case and the feature vector to the machine learning model, which outputs a predicted surgical case length. The prediction system reserves an operating room for the predicted surgical case length.

Abstract: Described herein are exemplary devices, apparatuses, systems, methods, and non-transitory storage media for providing an application framework. The application framework can provide various machine-learning models to perform a variety of analysis tasks to analyze enterprise data such as communications between one or more employees of an organization and one or more clients of the organization and provide intelligence and insights for a user in the organization. The insights and intelligence can include a recommendation or an observation related to a client or customer of the organization. The recommendation or observation can be provided, for example, in a communication platform, a chatbot, or a variety of other interfaces. Advantageously, to perform an analysis task, the application framework automatically provides to the machine-learning model(s) information in accordance with the enterprise's data sharing and access control requirements to prevent inappropriate access and use of sensitive information.

Abstract: One aspect of the present disclosure relates to a method of sentiment analysis based on ambiguity analysis, which includes analyzing information with the sentiment analysis models and the ambiguity analysis models. Another aspect of the present disclosure relates to a method of training the sentiment analysis models and ambiguity analysis models, which includes acquiring information, constructing lexicons, conducting sentiment analysis and ambiguity analysis with said lexicons, acquiring corpus, and training models, etc. Meanwhile, another aspect of the present disclosure relates to a system of sentiment analysis, which includes input, and output modules, acquisition modules, processing modules and database.

Abstract: Technology is disclosed herein for the integration of spreadsheet environments with an LLM service. In an implementation, an application receives a natural language input from a user associated with visualization of data hosted by the application. The application generates a prompt for a large language model (LLM) service based on the user input and the visualization and submits the prompt to the LLM service. The application receives a reply to the prompt from the LLM service and modifies the visualization based on the reply from the LLM service. In an implementation, the data includes spreadsheet data and the visualization includes a chart. In some implementations, to modify the visualization, the application generates source code based on the reply from the LLM service when the classification of the natural language input is a command.

Abstract: In one embodiment, a method is provided. The method includes obtaining sensor data indicative of a set of objects detected within an environment. The method also includes determining a set of positions of the set of objects and a set of properties of the set of objects based on the sensor data. The method further includes generating a state graph based on the sensor data. The state graph represents the set of objects and the set of positions of the set of objects. The state graph includes a set of object nodes to represent the set of objects and a set of property nodes to represent the set of properties of the set of objects. The state graph is provided to a graph enhancement module that updates the state graph with additional data to generate an enhanced state graph.

Abstract: A method implemented via execution of computing instructions configured to run at one or more processors and stored at one or more non-transitory computer-readable media. The method can include generating training data for an intent classification machine learning model by: (a) determining, via a text-to-text machine learning model, one or more respective paraphrases for each sample phrase of training phrases; (b) generating, via a label generating machine learning model, labeled data based on unlabeled live logs by: (i) determining live-log samples from the unlabeled live logs based at least in part on: a respective timestamp of each live log of the unlabeled live logs, or random sampling; and (ii) generating, via the label generating machine learning model, the labeled data based on the live-log samples and one or more labeling functions; and (c) adding the one or more respective paraphrases for the each sample phrase of the training phrases and the labeled data to the training data.

Abstract: Systems and methods for providing interactions of an Artificial Intelligence (AI) character model with users are provided. An example method includes receiving a message from a user of a client-side computing device; retrieving, from a graph, information concerning relationships between the AI character model and the user; generating, based on the message and the information concerning relationships, an action associated with AI character model; and causing the AI character model to perform the action in a virtual environment provided to the user via the client-side computing device. The client-side computing device may be in communication with a computing platform. The graph may include a first node associated with the AI character model, a second node associated with the user, and an edge between the first node and the second node. The edge may be associated with the information concerning relationships between the AI character model and the user.

Abstract: Methods and systems for efficient caching and retrieval of responses in conversation service applications includes a server that captures an utterance and converts the utterance into an utterance index key. The server searches a first response cache to determine whether the utterance index key matches a response index key. When there is a match, the server transmits a response that matches the utterance index key to a client device. When there is not a match, the server converts the utterance into an utterance embedding and searches a second response cache to identify a response embedding. The server captures a fuzzy response index key associated with the closest matching response embedding and searches the first response cache to identify a response index key that matches the fuzzy response index key.

Abstract: Methods, systems, and computer-readable media are disclosed herein for improved state identification and prediction in computerized queries. In an aspects, a neural network model is trained via word embedding, using a plurality of workflows having a plurality of steps as input training data. The model may be searched using a string to locate and identify semantic matches as potential results, where the potential results correspond to a specific step and/or a particular workflow. Markov chaining may also be performed, using the potential results, in order to predict one or more additional results, where the additional results correspond to a specific succeeding step within a particular workflow, in some aspects. The results and predicted steps may be displayed.

Abstract: A data processing system includes a queue manager receiving data processing requests and determining a queue depth representing the number of pending requests. A load supervisor assigns a service level to each request based on the queue depth when the request is at the head of the queue. The system offers two service levels, with the second level requiring fewer computing resources than the first. This dynamic management system optimizes resource allocation by adjusting service levels based on the workload, ensuring efficient processing of data requests.

Abstract: Certain aspects of the disclosure are directed to context aggregation in a data communications network. According to a specific example, user-data communications between a client-specific endpoint device and the other participating endpoint device during a first time period can be retrieved from a plurality of interconnected data communications systems. The client station can be configured and arranged to interface with a data communications server providing data communications services on a subscription basis. Context information for each respective user-data communication between the client station and the participating station during the first time period can be aggregated, such that subsequent user-data communications received from the participating station and intended for the client entity, can be routed based on the aggregated context information.

Abstract: This disclosure describes example implementations for generating context-dependent embedding vectors of words in the multi-dimensional embedding space based on a generic and non-domain-specific pretrained word embeddings. Such an implementation requires no domain specific training corpus and is capable of generating context-dependent embedding vectors of multi-semantic words using a few contextual texts. Such an implementation thus provides an efficient way to generate a library of multiple domain specific embedding vectors for multi-semantic words without any domain-specific training process. Other example embodiments further apply the principles of the context-dependent word embedding generation to a text-to-image application.