Abstract: A method and system of generalized balanced few-shot learning using only novel data without old data is proposed. The method of generalized balanced few-shot learning using only novel data without old data proposed in the present disclosure includes pre-training stage for training a feature extractor and a classifier of a training model with base data through a pre-training unit and fine-tuning stage for freezing the feature extractor through a fine-tuning unit, training a joint linear classifier capable of inferring base classes and novel classes, and performing weight normalization to achieve zero-mean and balanced variance.

Abstract: Certain aspects of the disclosure provide techniques for adaptive agentic workflow plan generation. A method generally includes determining that data, representing user interactions and previous system context that are associated with a graphical user interface (GUI), corresponds to a first navigation sequence, the first navigation sequence comprising first actions associated with the GUI for the previous system context; assigning a goal state to the first navigation sequence; generating a knowledge base comprising at least the first navigation sequence, the goal state, and an indication of first tools configured to perform the first actions; and generating, based on the knowledge base, an agentic workflow plan comprising second actions for execution by second tools to achieve the goal state, wherein the second actions and the second tools are dynamically adapted from the first actions and the first tools based on one or more factors.

Abstract: This disclosure describes systems that evaluating generative queries and modifying generative models (e.g., large language models or LLMs) and/or their system prompts based on the evaluation. The disclosed systems can generate and maintain a generative model functionality codex that stores tags for prompts and responses (to capture behaviors and intents of users and LLMs or other generative models). Using the codes, the disclosed systems can update evaluation model generation and can update queries and/or model parameters based on the evaluation. The disclosed systems can further distill models using updated queries and/or learned model parameters.

Abstract: Provided is a method of analyzing social influence. The method according to the embodiments includes obtaining posts containing text data in a forum, generating a forum interaction graph including a plurality of nodes corresponding to a plurality of embedded vectors generated by embedding the posts and edges indicating a connection relationship between the posts, and training an artificial intelligence model using the forum interaction graph.

Abstract: A system for automated model hyperparameter determination and serving includes an interface and a processor. The interface is configured to receive a request to determine optimal hyperparameters for a model. The processor is configured to determine a framework extension to implement the model; determine an optimization assembly to support the framework extension; determine a cluster to execute the optimization assembly; determine an optimization script for the optimization assembly; cause the cluster to execute the optimization script to determine the optimal hyperparameters; and build an optimal model using the optimal hyperparameters.

Abstract: Disclosed is a distributed computer system that includes a plurality of worker nodes that are coupled together via a data communication network to exchange data therebetween, wherein collective learning of the worker nodes is managed within the distributed computer system. The distributed computer system comprises a data processing arrangement operable to cluster the plurality of worker nodes into one or more clusters, wherein worker nodes of a given cluster train a computing model by employing a respective secondary distributed ledger. The collective learning from the plurality of worker nodes is coordinated using the distributed ledger arrangement.

Abstract: A transformer-based modeling architecture for training a generator to generate synthetic data. The architecture includes embedding models for embedding input data having multiple fields containing real data values and constructing individual embedding matrices for each of the multiple fields, a masking model for producing a set of masked fields in the embedded input dataset and replacing the real data value of each field with a mask token; and a transformer model for predicting the real data value for each mask token in each of the masked fields over multiple iterations.

Abstract: A system includes a computing platform having a hardware processor and a memory storing software code, a memory data structure storing memory features for an artificial intelligence interactive character (AIIC), and a trained machine learning (ML) model. The hardware processor executes the software code to receive interaction data describing a communication by a user with the AIIC, predict, using the trained ML model and the interaction data, at least one user memory feature(s) of the communication, and identify, using the memory data structure, one or more of the memory features for the AIIC as corresponding to the user memory feature(s). The software code also determines, using the user memory feature(s) and the corresponding one or more of the memory features for the AIIC, an interactive communication for execution by the AIIC in response to the communication by the user; and outputs the interactive communication to the AIIC.

Abstract: Aspects of the present disclosure involve efficient retrieval-augmented generation. Aspects include generating, by a first machine learning model based on a query and a knowledge corpus, a draft context comprising candidate knowledge items, and a first sufficiency score indicating a likelihood that the draft context is sufficient for the query. Aspects include generating, by a second machine learning model, based on the query and the draft context, a second sufficiency score indicating a corresponding likelihood that the draft context is sufficient for the query. Aspects include performing, based on the second sufficiency score and the first sufficiency score, one of: selecting the draft context as a generation context; or performing a comprehensive retrieval augmented generation process using the knowledge corpus to determine an alternative context for the query and selecting the alternative context as the generation context. Aspects include generating a response to the query based on the generation context.

Abstract: Systems, methods, and computer-readable storage media for generative artificial intelligence prompt and training data improvement are disclosed. Prompt and data improvement may be achieved through an iterative process. An example method by one or more processors may include receiving a first prompt for a generative artificial intelligence (AI) application; receiving one or more evaluation inputs; determining a performance of the first prompt based on the one or more evaluation inputs; and generating a second prompt based on the first prompt and the determined performance of the first prompt.

Abstract: A robotic surgical system network includes a plurality of robotic surgical systems. Each system includes robotic arms, sensors, a surgeon console, and a control system with an integrated AI module. A network interface is associated with robotic surgical system and provides secure data communication. A central or distributed data repository securely stores surgical data aggregated from the robotic surgical systems. The surgical data includes at least one of procedural data, sensor readings, imaging data, AI decision logs, surgical outcomes, or user interaction data. A training module utilizes aggregated surgical data to train or update AI models for the robotic surgical systems using unsupervised learning, transfer learning, or federated learning techniques. A cybersecurity module implements security measures for data transmission and system access, the measures comprising at least one of encryption, multi-factor authentication, or real-time threat detection.

Abstract: The present disclosure relates to a method and apparatus for sparsification training of a neural network model, a board card, and a readable storage medium. The data processing apparatus of the present disclosure is implemented as a computing apparatus and included in a combined processing apparatus. The combined processing apparatus further includes an interface apparatus and other processing apparatus. The computing apparatus interacts with other processing apparatus to jointly complete a computing operation specified by a user. The combined processing apparatus further includes a storage apparatus. The storage apparatus is connected to the computing apparatus and other processing apparatus, respectively. The storage apparatus is used to store data of the computing apparatus and other processing apparatus.

Abstract: Methods, apparatus, and computer readable storage medium for navigating multiple objects from initial positions towards target positions are described in the present disclosure.

Abstract: A system, method, and computer-readable medium for detecting a CRISPR-edited genome are disclosed. Certain embodiments of the system may include one or more processors configured to receive sequence data of a genome; generate an image representation of the sequenced data, the image being a plot of methylation variations as a function of methylation locations in the genome; apply the generated image representation to a trained convoluted neural network (CNN); generate, using the CNN, a score indicative of a probability that the genome was CRISPR-edited; and determine, based on the score, whether the genome contains a CRISPR-edited methylation region. A corresponding method and computer-readable medium are also provided.

Abstract: A system and method for determining device attributes based on protocol string conventions. A method includes applying at least one machine learning model to an application data set extracted based on at least one first pair of strings, each first pair of strings including a protocol string and a key string indicated in respective fields of communications session data corresponding to a device, wherein each machine learning model is trained based on a training data set including second pairs of strings device attribute labels, wherein each device attribute label corresponds to one of the second pairs of strings, wherein each of the at least one machine learning model is configured to output a predicted device attribute for the device based on the first pair of strings; and determining, based on the output of the at least one machine learning model, at least one device attribute of the device.

Abstract: A data management system including a processor connected to a storage device that stores data usable for learning and inference in a machine learning model, the data management system including a compressor/decompressor for human that compresses and decompresses data to be usable for verification performed by human, in which the processor specifies data that is no longer used for learning and inference in the machine learning model among the data stored in the storage device, and compresses the specified data using the compressor/decompressor for human.

Abstract: Systems, methods, apparatuses, and computer program products for neural networks. In accordance with some example embodiments, an operational neuron model may comprise an artificial neuron comprising a composite nodal operator, a pool-operator, and an activation function operator. The nodal operator may comprise a linear function or non-linear function. In accordance with certain example embodiments, a generative neuron model may include a composite nodal-operator generated during the training using Taylor polynomial approximation without restrictions. In accordance with various example embodiments, a self-organized operational neural network (Self-ONN) may include one or more layers of generative neurons.

Abstract: Methods and devices of a function approximation unit configured to approximate a nonlinear function within a neural processing unit are described. According to one embodiment, the method includes storing an input value through an input register of the function approximation unit, transmitting the input value to a selected one of a plurality of preprocessing circuits of the unit according to a control signal, generating a preprocessing result corresponding to the input value by the selected one of the preprocessing circuits, transmitting the preprocessing result to a programmable function approximation circuit and a selected one of a plurality of post-processing circuits of the unit, generating an approximated function output based on the preprocessing result in the programmable function approximation circuit, and generating a final output value by post-processing the preprocessing result or the approximated function output in the selected one of the post-processing circuits.

Abstract: Analog memory-based activation function for an artificial neural network can be provided. An apparatus can include at least two non-volatile memory devices connected in parallel such that the current can flow through one of the two non-volatile memory devices depending on the voltage level driving the current. To control which branch an input current flows through, each of the two non-volatile memory devices can be connected to a circuit element that can function as a switch, for example, a diode such as a semiconductor diode, a transistor, or another circuit element. Such apparatus can implement an analog memory-based activation function, for example, for an analog memory-based artificial neural network.

Abstract: A prediction model (1) includes a first module (M1) that calculates, for each of a plurality of objects (xi) in a dataset (x), an index value (vi) corresponding to a combination of the object (xi) and attribute information (a) using a neural network, and a second module (M2) that calculates a prediction result (y) of an operation to be performed by a user by performing a predetermined process on a plurality of index values (v1, . . . , vN) obtained from the first module (M1) and corresponding to the respective plurality of objects (x1, . . . , xN).