A GHeT includes a bottom gate formed on a substrate; a first dielectric layer (DL) formed on the bottom gate; a monolayer film formed of an atomically thin material on the first DL; a bottom contact (BC) formed on part of the monolayer film; a second DL formed on the BC; a top contact (TC) formed on the second DL on top of the BC; a network of CNTs formed on the TC and the monolayer film, to define an overlap region with the monolayer film; a third DL formed on the CNT network, the monolayer film and the TC; and a top gate formed on the third DL and overlapping with the overlap region. Such GHeT design allows gate tunability of Gaussian peak position, height and width that define Gaussian transfer characteristic, thereby enabling simplified circuit architectures for various spiking neuron functions for emerging neuromorphic applications.

Disclosed is a ray clustering learning method based on weakly-supervised learning for denoising using ray tracing. The ray clustering learning method is for learning a denoising model for removing noise from a rendered image through ray tracing, and includes extracting a feature of a simulated ray through the ray tracing and clustering the ray through contrastive learning for the feature.

Implementations disclosed describe methods and systems to perform the methods of optimizing a size of memory used for accumulation of neural node outputs and for supporting multiple computational paths in neural networks. In one example, a size of memory used to perform neural layer computations is reduced by performing nodal computations in multiple batches, followed by rescaling and accumulation of nodal outputs. In another example, execution of parallel branches of neural node computations include evaluating, prior to the actual execution, the amount of memory resources needed to execute a particular order of branches sequentially and select the order that minimizes this amount or keeps this amount below a target threshold.

Aspects described provide systems and methods that relate generally to image analysis and, more specifically, identifying individual components and elements in an image. The systems and methods include a valuation application executing one or more application program interfaces (APIs) communicating with one or more websites via a network, where the user is prompted to enter information and/or take pictures or videos of their vehicle that they would like to sell. The valuation application utilizes a machine learning model to identify and value the various vehicle components within the images and videos. Based on the machine learning model, the valuation application identifies each component according to the images and videos and performs a search to determine the value of the components identified. The valuation application tabulates and summarizes the vehicle component resale values and resell information for the user to view.

A method for providing a neural network system includes performing contrastive learning to the neural network system to generate a trained neural network system. The performing the contrastive learning includes performing first model augmentation to a first encoder of the neural network system to generate a first embedding of a sample, performing second model augmentation to the first encoder to generate a second embedding of the sample, and optimizing the first encoder using a contrastive loss based on the first embedding and the second embedding. The trained neural network system is provided to perform a task.

A system and method that given one or more input molecules, produces a contextualized summary of characteristics of related target molecules, e.g., proteins. Using a knowledge graph which is populated with all known molecules, input molecules are analyzed according to various similarity indexes which relate the input molecules to target proteins or other biological entities. The knowledge graph may also comprise scientific literature, governmental data (FDA clinical phase data), private research endeavors (general assays, etc.), and other related biological data. The summary produced may comprise target proteins that satisfy certain biological properties, general assay results (ADMET characteristics), related diseases, off-target molecule interactions (non-targeted molecules involved in a specific pathway or cascade), market opportunities, patents, experiments, and new hypothesis.

A computer system (which may include one or more computers) that trains a neural network is described. During operation, the computer system may train the neural network based at least in part on a set of hyperparameters, where the training includes computing weights associated with neurons in the neural network. Moreover, during the training, the computer system may dynamically adapt one or more first hyperparameters in the set of hyperparameters based at least in part on a measure corresponding to a local geometry of a loss landscape at or proximate to a current location in the loss landscape. Note that the dynamic adapting based at least in part on the measure is separate from or in addition to a predefined adaptation of one or more second hyperparameters the set of hyperparameters based on a predefined number of iterations or cycles in the training or a predefined scaling factor.

Herein are machine learning techniques that adjust reconstruction loss of a reconstructive model, such as a principal component analysis (PCA), based on importances of features. In an embodiment having a reconstructive model that more or less accurately reconstructs its input, a computer measures, for each feature, a respective importance that is based on the reconstructive model. For example, importance may be based on grading samples that the reconstructive model correctly or incorrectly inferenced. For each feature during production inferencing, a respective original loss from the reconstructive model measures a difference between a value of the feature in an input and a reconstructed value of the feature generated by the reconstructive model. For each feature, the respective importance of the feature is applied to the respective original loss to generate a respective weighted loss, which compensates for concept drift. The weighted losses of the features of the input are collectively detected as anomalous or non-anomalous.

One or more computing devices, systems, and/or methods for cross-domain action prediction are provided. Action sequence embeddings are generated based upon a textual embedding and a graph embedding utilizing past user action sequences corresponding to sequences of past actions performed by users across a plurality of domains. An autoencoder is trained to utilize the action sequence embeddings to project the action sequence embeddings to obtain intent space vectors. A service switch classifier is trained using the intent space vectors. In response to the service switch classifier predicting that a current user will switch from a current domain to a next domain, the current user is provided with a recommendation of an action corresponding to the next domain.

A method includes: partitioning learning data containing objective variables and explanatory variables into a plurality of subsets of data; executing regularization processing on first data in each of the partitioned subsets, and extracting a first element equal to zero; extracting, as a candidate, each model where an error ratio between first multiple regression and second multiple regression is equal to or more than a predetermined value, the first multiple regression being a result of multiple regression on second data which is test data in each of the partitioned subsets and is for use to calculate the error ratio of the learning data, the second multiple regression being a result of multiple regression on third data obtained by excluding the first element from the second data; and outputting a model where zero is substituted for an element that takes zero a predetermined or larger number of times in the candidate.