Abstract: Semantic frame identification involves associating identified target words in the sentential context of their natural language source with semantic frames from a frame lexical database. The disclosed invention leverages the CapsNet architecture for improved semantic frame identification of a target word in a natural language input. This includes deriving the features of a target word identified in the sentence and extracting the features of the word units and the thematic words around the target word. Through dynamic routing of capsules, the CapsNet is able to filter the candidate frames for the target word to reduce the search space and apply the CapsNet prediction to identify a frame from a frame lexical database.

Abstract: A method and system for persisting data are provided. Batch data is periodically extracted via a computer system from at least one primary data source. Batch data is transformed via the computer system. The batch data is loaded, via the computer system, into a database. Real-time data is continuously received from at least one secondary data source. The received real-time data is indexed in an index. Business intelligence reports are generated from both the real-time data in the index and the batch data in the database.

Abstract: Embodiments described herein provide for training an artificial intelligence model to become a preference-aware model. The artificial intelligence model preferences as the artificial intelligence model trains. Reinforcement learning is used to train experts in the artificial intelligence model such that each expert is trained to converge to a unique preference. The architecture of the artificial intelligence model is highly flexible. Upon executing a trained model, users can select automatically images according to various preferences based on medical professional preferences, geographic preferences, patient anatomy, and institutional guidelines.

Abstract: A method for detecting slow HTTP DoS (SHD) attacks in a backbone network can detect three different types of SHID attacks. The method is divided into an off-line training phase and an on-line detection phase. In the off-line training phase, several types of representative unidirectional traffic features are extracted according to attack characteristics of different SHD types and corresponding feature groups are built, where these features can effectively deal with a large amount of unidirectional traffic in backbone networks; a public backbone network dataset is systematically sampled and data are stored in combination with Count-min Sketch, which greatly minimizes storage and computational overhead required in the backbone networks; and a specific machine learning algorithm is used for training to obtain attack detection models. The method can be used for detecting and warning SHD attacks in mass traffic scenarios such as backbone networks to provide a basis for maintaining network security.

Abstract: Provided are systems and methods for generating a score for any model which can be updated online, regardless of model type architecture and parameters, leveraging relations between regret and uncertainty.

Abstract: Aspects are provided which allow a UE to achieve sidelink parameter coordination through various signaling in clustered FL or peer-to-peer FL. The UE provides a first ML model information update and a measurement associated with the update. The UE obtains an aggregated ML model information update aggregating the first update with a second ML model information update of either a first network node in a FL cluster including the UE or a second network node in a second FL cluster. The UE provides a sidelink communication parameter in SCI to a network node, which parameter is an output of an ML model associated with the aggregated ML model information update. As a result, UEs may derive common SCI parameters from their updated ML models to apply in sidelink communications, thereby leading to maximized packet reception rate, maximized throughput, or minimized latency in communication.

Abstract: A method includes generating a plurality of binary feature maps containing a set of feature map values including a first binary value and/or a second binary value, by at least converting each input value of a set of input values of a plurality of input feature vectors to the first binary value when the corresponding input value is the zero value or the second binary value when the corresponding input value is the non-zero value. The method includes segmenting the plurality of binary feature maps into a plurality of segments representing behavior profiles. Each segment includes at least one subsegment in which the set of feature map values is the same for all binary feature maps in the at least one subsegment. The method includes predicting, based on a segment of the plurality of segments, a specific outcome. Related methods and articles of manufacture are also disclosed.

Abstract: Methods, systems, and apparatus for selecting actions to be performed by an agent interacting with an environment. One system includes a high-level controller neural network, low-level controller network, and subsystem. The high-level controller neural network receives an input observation and processes the input observation to generate a high-level output defining a control signal for the low-level controller. The low-level controller neural network receives a designated component of an input observation and processes the designated component and an input control signal to generate a low-level output that defines an action to be performed by the agent in response to the input observation.

Abstract: A processor comprising a barrel-threaded execution unit for executing concurrent threads, and one or more register files comprising a respective set of context registers for each concurrent thread. One of the one or more register files further comprises a set of shared weights registers common to some or all of the concurrent threads. The types of instructions defined in the instruction set of the processor include an arithmetic instruction having operands specifying a source and a destination from amongst a respective set of arithmetic registers of the thread in which the arithmetic instruction is executed. The execution unit is configured so as, in response to the opcode of the arithmetic instruction, to perform an operation comprising multiplying an input from the source by at least one of the weights from at least one of the shared weights registers, and to place a result in the destination.

Abstract: A computer-implemented method for maintaining a function of a local entity upon connection disruption to a backend in a communication system including a backend and a plurality of local entities. The backend provides backend information for the function. The method includes reception of local behavior models from a plurality of local entities by the backend, wherein the local behavior models provide the function in the local entities if particular backend information is not available; creation of a behavior model based on the received local behavior models; and transmission of the behavior model to a local entity of the plurality of the local entities by the backend.

Abstract: A method, a computer program product, and a computer system classify an image with a convolutional neural network. The method receiving an image. The method includes performing a radial summation on the image to generate a radially summed image. The method includes inputting the radially summed image into the CNN to perform an image classification.

Abstract: Bias in Machine Learning (ML) is when an ML algorithm tends to incompletely learn relevant and important patterns from a dataset, or learns the patterns from data incorrectly. Such inaccuracy can cause the algorithm to miss important relationships between patterns and features in data, resulting in inaccurate algorithm predictions. Systems and methods for detecting potential ML bias in input image datasets are described herein. After a target image is received, a subset of images related to the target image is extracted. The target image and subset of images are analyzed under an imbalance assessment and data bias assessment to determine the presence of any potential data bias in a ML training pipeline. If any data bias is determined, one or more messages summarizing the assessments and including explanations to enable more accurate predictions in image assessments are sent to the user.

Abstract: A method for operating a terminal with high availability in a network, where a memory stores an identifier, for a backup device of a first edge device, which relates to a second edge device, stores a model for operating the terminal and stores information about the connection configuration between the first and second edge devices, the monitoring device and the terminal, where the first edge device periodically sends a signal about its correct operation, the signal is received by the monitoring device, and detection of an absent or erroneous signal results in an associated error device being detected, and the model and the connection configuration of the error device being transferred from the memory to a substitute device, the substitute device being determined via the identifier associated with the absent or erroneous signal, and the substitute device operating the terminal using the received model and the connection configuration.

Abstract: In an example, an apparatus comprises a compute engine comprising a high precision component and a low precision component; and logic, at least partially including hardware logic, to receive instructions in the compute engine; select at least one of the high precision component or the low precision component to execute the instructions; and apply a gate to at least one of the high precision component or the low precision component to execute the instructions. Other embodiments are also disclosed and claimed.

Abstract: A neural network control system for controlling an agent to perform a task in a real-world environment, operates based on both image data and proprioceptive data describing the configuration of the agent. The training of the control system includes both imitation learning, using datasets generated from previous performances of the task, and reinforcement learning, based on rewards calculated from control data output by the control system.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for data-efficient reinforcement learning. One of the systems is a system for training an actor neural network used to select actions to be performed by an agent that interacts with an environment by receiving observations characterizing states of the environment and, in response to each observation, performing an action selected from a continuous space of possible actions, wherein the actor neural network maps observations to next actions in accordance with values of parameters of the actor neural network, and wherein the system comprises: a plurality of workers, wherein each worker is configured to operate independently of each other worker, wherein each worker is associated with a respective agent replica that interacts with a respective replica of the environment during the training of the actor neural network.

Abstract: Devices, methods and computer programs for a transformer-based decoder for channel state information compression are disclosed. At least some example embodiments may allow at least some of the example embodiments described herein may allow a lightweight general purpose, high-fidelity, and data-driven decoder for channel state information (CSI) array compression associated with massive multiple-input and multiple-output (MIMO).

Abstract: A system and method are disclosed for providing a private multi-modal artificial intelligence platform. The method includes splitting a neural network into a first client-side network, a second client-side network and a server-side network and sending the first client-side network to a first client. The first client-side network processes first data from the first client, the first data having a first type. The method includes sending the second client-side network to a second client. The second client-side network processes second data from the second client, the second data having a second type. The first type and the second type have a common association. Forward and back propagation occurs between the client side networks and disparate data types on the different client side networks and the server-side network to train the neural network.

Abstract: Disclosed are a system, method and apparatus to generate service codes based, at least in part, on electronic documents.

Abstract: Systems, methods, and computer program products are provided for generating and improving multitask learning models. An example method includes determining a first accuracy metric based on at least two machine learning models performing a plurality of tasks, receiving a multitask learning model including at least one shared layer and a plurality of task-specific layers, determining a second accuracy metric based on the multitask learning model having a first number of shared layers, determining a third accuracy metric based on the multitask learning model having a second number of shared layers, comparing the accuracy metrics, repeating until at least one termination condition is satisfied, and determining a target number of shared layers for the multitask learning model based on at least one of the second accuracy metric, the third accuracy metric, the first number of shared layers, the second number of shared layers, or any combination thereof.

Abstract: Examples relate to an edge controller apparatus, to an edge controller method, to an edge controller computer program, and to systems comprising an edge controller apparatus. The edge controller apparatus comprises at least one interface for obtaining status data of a plurality of sensors or actuators and for communicating with a remote server. The edge controller apparatus comprises processing circuitry configured to obtain a plurality of units of status data of the plurality of sensors or actuators, the plurality of sensors or actuators being part of machinery or being arranged in an environment of the machinery. The processing circuitry is configured to update at least one model representing the status of the sensors or actuators based on the plurality of units of status data. The processing circuitry is configured to provide the updated at least one model representing the status of the sensors or actuators to the remote server.

Abstract: A storage and computation integrated apparatus and a calibration method therefor. The storage and computation integrated apparatus includes a first processing unit, which includes: a first computation memristor array; a first calibration memristor array; and a first processing unit. The calibration method includes: determining, by means of off-chip training, a first computation weight matrix which corresponds to a first computation memristor array, and writing the first computation weight matrix into the first computation memristor array; and on the basis of the first computation memristor array where the first computation weight matrix has been written and the first computation weight matrix, performing on-chip training on a first calibration memristor array, so as to adjust a weight value of the first calibration memristor array.

Abstract: According to one embodiment, an arithmetic apparatus includes a non-volatile first memory, a second memory, and a controller. The first memory stores a model to be trained. The second memory has a smaller storage capacity than the first memory. The controller executes learning processing that updates a first parameter of the model based on a loss value obtained by inputting training data into the model stored in the first memory, and stores cumulative update information indicating a difference of the first parameter before and after the update in the second memory. In addition, the controller executes the learning processing using a second parameter in which the cumulative update information stored in the second memory is reflected in the first parameter read from the model stored in the first memory, and stores a difference between a third parameter obtained by updating the second parameter and the first parameter, in the second memory as the cumulative update information.

Abstract: A system and method for training a neural network. In some embodiments, the system includes a computational storage device including a backing store. The computational storage device may be configured to: store, in the backing store, an embedding table for a neural network embedding operation; receive a first index vector including a first index and a second index; retrieve, from the backing store: a first row of the embedding table, corresponding to the first index, and a second row of the embedding table, corresponding to the second index; and calculate a first embedded vector based on the first row and the second row.

Abstract: Systems, devices, and methods related to a Deep Learning Accelerator and memory are described. For example, an integrated circuit device may be configured to execute instructions with matrix operands and configured with random access memory. The random access memory is configured to store input data from a sensor, parameters of a first portion of an Artificial Neural Network (ANN), instructions executable by the Deep Learning Accelerator to perform matrix computation of the first portion of the ANN, and data generated outside of the device according to a second portion of the ANN. The Deep Learning Accelerator may execute the instructions to generate, independent of the data from the second portion of the ANN, a first output based on the input data from the sensor and generate a second output based on a combination of the data from the sensor and the data from the second portion of the ANN.

Abstract: In an approach for managing data continuity of a data storage system, a processor analyzes queries received by the data storage system from a plurality of data sources. A processor identifies a query frequency and response time for queries from each data source of the plurality of data sources. A processor derives an expected data pipeline continuity requirement for each data source based on respective query frequency and respective response time for the respective queries from each data source. A processor learns patterns of the analyzed queries to enable predictions for when future queries from each data source will occur and identification of an unavailability pattern for each data source. A processor determines recommendations for maintaining data continuity for the data storage system based on the expected data pipeline continuity requirement for each data source, the query predictions, and the unavailability pattern for each data source.

Abstract: In one aspect, a method of training a DNN includes transmitting an input vector x through a weight matrix W and reading a resulting output vector y, transmitting an error signal ?, transmitting the input vector x with the error signal ? through conductive row wires of a matrix A, and transmitting an input vector ei and reading a resulting output vector y? as current output. The training also includes updating a hidden matrix H comprising an H value for RPU devices by iteratively adding the output vector y? multiplied by the transpose of the input vector ei to each H value. The training also includes, when an H value reaches a threshold value, transmitting the input vector ei as a voltage pulse through the conductive column wires of the matrix W simultaneously with sign information of the H values that reached a threshold value as voltage pulses through the conductive row wires matrix W.

Abstract: A neural network system is configured to receive an input image and to generate a classification output for the input image. The neural network system includes: a separable convolution subnetwork comprising a plurality of separable convolutional neural network layers arranged in a stack one after the other, in which each separable convolutional neural network layer is configured to: separately apply both a depthwise convolution and a pointwise convolution during processing of an input to the separable convolutional neural network layer to generate a layer output.

Abstract: A method for deep neural network compression is provided. The method includes: using at least one weight of a deep neural network (DNN), setting a value of a P parameter, and combining every P weights in groups, and perform branch pruning and retraining, so that only one of each group has a non-zero weight, and the remaining weights are 0, wherein the remaining weights are evenly divided into branches to adjust a compression rate of the DNN and to adjust a reduction rate of the DNN.

Abstract: There is provided a voltage control device that automatically sets a limit operating voltage. Provided is a voltage control device including a first neural network, a second neural network, an inference result determination unit, and a voltage determination unit, in which the inference result determination unit has a function of comparing correct answer value data held by the inference result determination unit with inference result data of the first neural network to obtain determination result data, and the voltage determination unit has a function of outputting a voltage signal lower than a voltage supplied to the first neural network and the second neural network in a case where the correct answer value data and the inference result data match, and outputting a voltage signal higher than the voltage supplied to the first neural network and the second neural network in a case where the correct answer value data and the inference result data do not match, on the basis of the determination result data.

Abstract: Images classified as including defects can be analyzed using localized anomaly detection to determine which regions of the images contributed to the classification. These techniques may identify one or more regions of the images having a confidence value exceeding a threshold or range as a factor in the classification. Using a number of different localized anomaly detection techniques may enable aggregation of the results to determine regions that overlap between the different techniques, thereby providing a higher confidence that these regions include defects. Such an approach can not only identify whether a defect is present, but also identify locations, sizes, or types of defects.

Abstract: A method for detection and characterization of lesions includes acquiring a plurality of phase images of a multi-phase imaging exam, extracting a local context for each phase image of the plurality of phase images, encoding the local contexts to create phase specific feature maps, combining the phase-specific feature maps to create unified feature maps, and at least one of characterizing or detecting a lesion based on the unified feature maps.

Abstract: A method and system for multiple-input multiple-output (MIMO) detector selection using a neural network is herein disclosed. According to one embodiment, a method includes generating a labelled dataset of features and detector labels, training a multi-layer perceptron (MLP) network using the generated labelled dataset, and selecting a detector class from a plurality of detector classes based on outputs of the trained MLP network.

Abstract: Embodiments disclosed herein include a method of training a DNN. A processor initializes an element of an A matrix. The element may include a resistive processing unit. A processor determines incremental weight updates by updating the element with activation values and error values from a weight matrix multiplied by a chopper value. A processor reads an update voltage from the element. A processor determines a chopper product by multiplying the update voltage by the chopper value. A processor directs storage of an element of a hidden matrix. The element of the hidden matrix may include a summation of continuous iterations of the chopper product. A processor updates a corresponding element of a weight matrix based on the element of the hidden matrix reaching a threshold state.

Abstract: A process for optimizing loss functions includes progressively building better sets of parameters for loss functions represented as multivariate Taylor expansions in accordance with an iterative process. The optimization process is built upon CMA-ES. At each generation (i.e., each CMA-ES iteration), a new set of candidate parameter vectors is sampled. These candidate parameter vectors are sampled from a multivariate Gaussian distribution representation that is modeled by the CMA-ES covariance matrix and the current mean vector. The candidates are then each evaluated by training a model (neural network) using the candidates and determining a fitness value for each candidate against a validation data set.

Abstract: Methods and systems for beam search decoding. One of the methods includes initializing beam data specifying a set of k candidate output sequences and a respective total score for each of the candidate output sequences; updating the beam data at each of a plurality of decoding steps, comprising, at each decoding step: generating a score distribution that comprises a respective score for each token in the vocabulary; identifying a plurality of expanded sequences; generating, for each expanded sequence, a respective backwards-looking score; generating, for each expanded sequence, a respective forward-looking score; computing, for each expanded sequence, a respective total score from the respective forward-looking score for the expanded sequence and the respective backwards-looking score for the expanded sequence; and updating the set of k candidate output sequences using the respective total scores for the expanded sequences.

Abstract: Systems and methods for firmware-based diagnostics in heterogenous computing platforms are described. In an illustrative, non-limiting embodiment, an Information Handling System (IHS) may include a heterogeneous computing platform having a plurality of devices and a memory coupled to the platform, where the memory includes firmware instructions that, upon execution by a respective device among the plurality of devices, enables the respective device to provide a corresponding firmware service, and wherein at least one of the plurality of devices operates as an orchestrator configured to: execute or instruct a selected device among the plurality of devices to execute an Artificial Intelligence (AI) model configured to determine whether to trigger a diagnostics process; and, in response to the determination, trigger the diagnostics process.

Abstract: A method for generating an object includes: providing a dataset having object data and condition data; processing the object data to obtain latent object data and latent object-condition data; processing the condition data to obtain latent condition data and latent condition-object data; processing the latent object data and the latent object-condition data to obtain generated object data; processing the latent condition data and latent condition-object data to obtain generated condition data; comparing the latent object-condition data to the latent condition-object data to determine a difference; processing the latent object data and latent condition data and one of the latent object-condition data or latent condition-object data to obtain a discriminator value; and selecting a selected object based on the generated object data.

Abstract: A method including training, by one or more processors, a generative AI model using first operating data from building equipment and a plurality of first service reports indicating a plurality of first problems associated with the building equipment. The method may include predicting, by the one or more processors using the generative AI model, one or more future problems likely to occur with the building equipment based on second operating data from the building equipment. The method may include automatically initiating, by the one or more processors, one or more actions to prevent the one or more future problems from occurring or mitigate an effect of the one or more future problems.

Abstract: A method, computer system, and a computer program product for invariant risk minimization games is provided. The present invention may include defining a plurality of environment-specific classifiers corresponding to a plurality of environments. The present invention may also include constructing an ensemble classifier associated with the plurality of environment-specific classifiers. The present invention may further include initiating a game including a plurality of players corresponding to the plurality of environments. The present invention may also include calculating a nash equilibrium of the initiated game. The present invention may further include determining an ensemble predictor based on the calculated nash equilibrium. The present invention may include deploying the determined ensemble predictor associated with the calculated nash equilibrium to make predictions in a new environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an action selection neural network used to select actions to be performed by an agent interacting with an environment. One of the systems includes (i) a plurality of actor computing units, in which each of the actor computing units is configured to maintain a respective replica of the action selection neural network and to perform a plurality of actor operations, and (ii) one or more learner computing units, in which each of the one or more learner computing units is configured to perform a plurality of learner operations.

Abstract: Embodiments of the present disclosure relate to a tensor access operation circuit in a neural processor circuit. The neural processor circuit further includes a data processor circuit and at least one neural engine circuit. The tensor access operation circuit indirectly accesses at least a region of a source tensor in a system memory having a rank, and maps one or more source components of the source tensor into an input tensor having another rank. The data processor circuit stores an output version of the input tensor obtained from the tensor access operation circuit and sends the output version of the input tensor as multiple of units of input data to the at least one neural engine circuit. The at least one neural engine circuit performs at least convolution operations on the units of input data and at least one kernel to generate output data.

Abstract: The present disclosure proposes the use of a dual discriminator network that comprises a temporal discriminator network for discriminating based on temporal features of a series of images and a spatial discriminator network for discriminating based on spatial features of individual images. The training methods described herein provide improvements in computational efficiency. This is achieved by applying the spatial discriminator network to a set of one or more images that have reduced temporal resolution and applying the temporal discriminator network to a set of images that have reduced spatial resolution. This allows each of the discriminator networks to be applied more efficiently in order to produce a discriminator score for use in training the generator, whilst maintaining accuracy of the discriminator network. In addition, this allows a generator network to be trained to more accurately generate sequences of images, through the use of the improved discriminator.

Abstract: Disclosed are a method, an apparatus, a device, and a medium for image recognition via wireless federated learning. The method for image recognition via wireless federated learning includes: obtaining an image to be recognized and an initial image recognition model; adjusting parameters for the initial image recognition model via a preset accelerated mobile federated learning algorithm according to a target momentum factor to obtain a target image recognition model; and recognizing the image to be recognized through the target image recognition model to obtain text information corresponding to the image to be recognized.

Abstract: Methods, systems, and computer program products for factchecking artificial intelligence models using blockchain are provided herein. A computer-implemented method includes obtaining at least one artificial intelligence model and at least one set of data related to the at least one artificial intelligence model; determining a set of characteristics based at least in part on the at least one artificial intelligence model and the at least one set of data; selecting one of a plurality of networks based at least in part on a target deployment of the at least one artificial intelligence model to verify the set of characteristics; generating a report based at least in part on verifying the set of characteristics using the selected network, wherein the report establishes a threshold level of trust for the at least one artificial intelligence model; and storing the report on a blockchain.

Abstract: Systems and methods for generating synthetic intercorrelated data are disclosed. For example, a system may include at least one memory storing instructions and one or more processors configured to execute the instructions to perform operations. The operations may include training a parent model by iteratively performing steps. The steps may include generating, using the parent model, first latent-space data and second latent-space data. The steps may include generating, using a first child model, first synthetic data based on the first latent-space data, and generating, using a second child model, second synthetic data based on the second latent-space data. The steps may include comparing the first synthetic data and second synthetic data to training data. The steps may include adjusting a parameter of the parent model based on the comparison or terminating training of the parent model based on the comparison.

Abstract: A method of determining a counterfeit fingerprint by a system for determining a counterfeit fingerprint that includes an internal light source and an external light source, comprising: extracting a first fingerprint area of a first fingerprint image obtained from a light signal of the internal light source when a target object's fingerprint comes in contact with a fingerprint contact surface of the system for determining a counterfeit fingerprint; extracting a second fingerprint area of a second fingerprint image obtained from a light signal of the external light source based on the first fingerprint area; and inputting the first fingerprint area and the second fingerprint area into a pre-trained neural network of the system for determining a counterfeit fingerprint to output a result of determining whether the fingerprint is counterfeit.

Abstract: A computer-implemented method for reinforcement learning with Logical Neural Networks (LNNs) is provided including receiving a plurality of observation text sentences from a target environment, extracting one or more propositional logic values from the plurality of observation text sentences, finding a class for each propositional logic value by using external knowledge, converting each propositional logic value into a first-order logic by replacing a part in the propositional logic value with a variable word, the part indicating the class, selecting a LNN based on the class among LNNs prepared in advance for each class, each LNN receiving the one or more propositional logic values as a status input and outputting an action with a score indicating a degree of preference for taking the action, and performing a highest score action to the target environment to obtain a next state of the target environment and a reward for the highest score action.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing inputs using a neural network system that includes one or more pre-normalized layers or one or more regularization normalization layers.

Abstract: The present disclosure a method of operating user equipment (UE) in a wireless communication system, the method comprising: identifying layer information that is applied to a neural polar code; generating, based on the identified layer information, transmission data by encoding data that is input into the neural polar code; and transmitting the transmission data to a base station, wherein, based on polar code transformation, the neural polar code generates the transmission data by performing encoding, based on the polar code transformation, from an initial layer of the data to a first layer according to the identified layer information and by performing encoding through a neural network-based autoencoder after the first layer until the transmission data is generated.