Certain aspects of the present disclosure provide techniques for performing machine learning computations in a compute in memory (CIM) array comprising a plurality of bit cells, including: determining that a sparsity of input data to a machine learning model exceeds an input data sparsity threshold; disabling one or more bit cells in the CIM array based on the sparsity of the input data prior to processing the input data; processing the input data with bit cells not disabled in the CIM array to generate an output value; applying a compensation to the output value based on the sparsity to generate a compensated output value; and outputting the compensated output value.

A computer-implemented neural network graph (1) system, comprising a plurality of neurons (2), each represented by a unique addressable node in a dynamic data structure and each containing a plurality of data, and a plurality of axons and dendrites (4) connecting two or more neurons (2) between them in order to represent a relation and transport one or more data contained in a neuron (2) to another neuron. Each axon (4) having at its end a synapse (3) for connecting it to a neuron (2) and at least one intermediate neuron (2) is connected through an intermediate axon (4) or dendrite and its synapse (3) directly to another axon (4) which connects two main neurons (2). The intermediate neuron (2) and intermediate axon (4) being configured for: selecting one or more specific data contained in the main neurons (2) and transmitted between them along their axon (4) or dendrites (4) in function of a preselected data of the intermediate neuron (2) in such a way to define a first combination of data; selecting one or more specific data, different from the first selection, contained in the main neurons (2) and transmitted between them along the axon (4) in function of a preselected data of the intermediate neuron (2) in such a way to define a second combination of data different from the first; creating a graphical representation comprising a graph (1) of said data in which a first abstraction level is defined by said first selection and a second abstraction level is defined by said second selection different from the first.

This disclosure relates to recommendations made to users based on learned behavior patterns. User behavior data is collected and grouped according labels. The grouped user behavior data is labeled and used to train a machine learning model based on features and tasks associated with the classification. User behavior is then predicted by applying the trained machine learning model to the collected user behavior data, and a task is recommended to the user.

A method of performing a reshape operation specified in a reshape layer of a neural network model is described. The reshape operation reshapes an input tensor with an input tensor shape to an output tensor with an output tensor shape. The tensor data that has to be reshaped is directly routed between tile memories of the hardware accelerator in an efficient manner. This advantageously optimizes usage of memory space and allows any number and type of neural network models to be run on the hardware accelerator.

The present disclosure relates to neural network optimization methods and apparatuses in the field of artificial intelligence. One example method includes sampling preset hyperparameter search space to obtain multiple hyperparameter combinations. Multiple iterative evaluations are performed on the multiple hyperparameter combinations to obtain multiple performance results of each hyperparameter combination. Any iterative evaluation comprises obtaining at least one performance result of each hyperparameter combination, and if a hyperparameter combination meets a first preset condition, re-evaluating the hyperparameter combination to obtain a re-evaluated performance result of the hyperparameter combination. An optimal hyperparameter combination is determined. If the optimal hyperparameter combination does not meet a second preset condition, a preset model is updated, based on the multiple performance results of each hyperparameter combination, for next sampling. Or if the optimal hyperparameter combination meets a second preset condition, the optimal hyperparameter combination is used as a hyperparameter combination of a neural network.

A method of training a deep learning model, a method of processing a natural language, an electronic device, and a storage medium are provided, which relate to a field of artificial intelligence, in particular to deep learning technology and natural language processing technology. The method includes: inputting first sample data into a first deep learning model to obtain a first output result; training the first deep learning model according to the first output result and a first target output result, the first target output result is obtained by processing the first sample data using a reference deep learning model; inputting second sample data into a second deep learning model to obtain a second output result; and training the second deep learning model according to the second output result and a second target output result, to obtain a trained second deep learning model.

Disclosed are a method and system for changing a structure of a deep learning model based on a change in resolution of input data. The method of changing a structure of a deep learning model may include generating, by the at least one processor, a plurality of input data having different resolution by performing various resolution changes on input data having given resolution, performing, by the at least one processor, inference on each of the plurality of generated input data through a deep learning model, checking, by the at least one processor, the size of a feature map output by each of layers included in the deep learning model while the inference is performed, and changing, by the at least one processor, the structure of at least one of the layers based on the checked size of the feature map.

A neural network pruning system can sparsely prune neural network models using an optimizer based approach that is agnostic to the model architecture being pruned. The neural network pruning system can prune by operating on the parameter vector of the full model and the gradient vector of the loss function with respect to the model parameters. The neural network pruning system can iteratively update parameters based on the gradients, while zeroing out as many parameters as possible based a preconfigured penalty.

A neural network pruning system can sparsely prune neural network models using an optimizer based approach that is agnostic to the model architecture being pruned. The neural network pruning system can prune by operating on the parameter vector of the full model and the gradient vector of the loss function with respect to the model parameters. The neural network pruning system can iteratively update parameters based on the gradients, while zeroing out as many parameters as possible based a preconfigured penalty.

Machine learning models used in medical diagnosis should be validated after being deployed in order to reduce the number of misdiagnoses. Validation processes presented here assess a performance of the machine learning model post-deployment. In post-deployment validation, the validation process monitoring can include: (1) monitoring to ensure a model performs as well as a reference member such as another machine learning model, and (2) monitoring to detect anomalies in data. This post-deployment validation helps identify low-performing models that are already deployed, so that relevant parties can quickly take action to improve either the machine learning model or the input data.