A method for verifying a calculation of a neuron value of multiple neurons of a neural network, including: carrying out or triggering a calculation of neuron functions of the multiple neurons, in each case to obtain a neuron value, the neuron functions being determined by individual weightings for each neuron input; calculating a first comparison value as the sum of the neuron values of the multiple neurons; carrying out or triggering a control calculation with one or multiple control neuron functions and with all neuron inputs of the multiple neurons, to obtain a second comparison value as a function of the neuron inputs of the multiple neurons and of the sum of the weightings of the multiple neurons assigned to the respective neuron input; and recognizing an error as a function of the first comparison value and of the second comparison value.

A stuck-at fault mitigation method for resistive random access memory (ReRAM)-based deep learning accelerators, includes: confirming a distorted output value (Y0) due to a stuck-at fault (SAF) by using a correction data set in a pre-trained deep learning network, by means of ReRAM-based deep learning accelerator hardware; updating an average (μ) and a standard deviation (σ) of a batch normalization (BN) layer by using the distorted output value (Y0), by means of the ReRAM-based deep learning accelerator hardware; folding the batch normalization (BN) layer in which the average (μ) and the standard deviation (σ) are updated into a convolution layer or a fully-connected layer, by means of the ReRAM-based deep learning accelerator hardware; and deriving a normal output value (Y1) by using the deep learning network in which the batch normalization (BN) layer is folded, by means of the ReRAM-based deep learning accelerator hardware.

Herein is machine learning for anomalous graph detection based on graph embedding, shuffling, comparison, and unsupervised training techniques that can characterize an unfamiliar graph. In an embodiment, a computer obtains many known vectors that respectively represent known graphs. A new vector is generated that represents a new graph that contains multiple vertices. The new vector may contain an arithmetic aggregation of vertex vectors that respectively represent multiple vertices and/or a vector that represents a virtual vertex that is connected to the multiple vertices by respective virtual edges. In the many known vectors, some similar vectors that are similar to the new vector are identified. The new graph is automatically characterized based on a subset of the known graphs that the similar vectors represent.

Computer systems and associated methods are disclosed to implement a model development environment (MDE) that allows a team of users to perform iterative model experiments to develop machine learning (ML) media models. In embodiments, the MDE implements a media data management interface that allows users to annotate and manage training data for models. In embodiments, the MDE implements a model experimentation interface that allows users to configure and run model experiments, which include a training run and a test run of a model. In embodiments, the MDE implements a model diagnosis interface that displays the model's performance metrics and allows users to visually inspect media samples that were used during the model experiment to determine corrective actions to improve model performance for later iterations of experiments. In embodiments, the MDE allows different types of users to collaborate on a series of model experiments to build an optimal media model.

Remote access for debugging or profiling a remotely executing neural network graph can be performed by a client using an in-band application programming interface (API). The client can provide indicator flags for debugging or profiling in an inference request sent to a remote server computer executing the neural network graph using the API. The remote server computer can collect metadata for debugging or profiling during the inference operation using the neural network graph and send it back to the client using the same API. Additionally, the metadata can be collected at various granularity levels also specified in the inference request.

A method of performing fault detection during computations relating to a neural network comprising a first neural network layer and a second neural network layer in a data processing system, the method comprising: scheduling computations onto data processing resources for the execution of the first neural network layer and the second neural network layer, wherein the scheduling includes: for a given one of the first neural network layer and the second neural network layer, scheduling a respective given one of a first computation and a second computation as a non-duplicated computation, in which the given computation is at least initially scheduled to be performed only once during the execution of the given neural network layer; and for the other of the first and second neural network layers, scheduling the respective other of the first and second computations as a duplicated computation.

Machine learning of model parameters for a neural network using a computing system is provided, that produces error-aware model parameters. An iterative process to converge on trained model parameters to be applied in the inference engine, includes applying a sequence of input training data sets to a neural network to produce inference results for the sequence using a set of model parameters in the neural network combined with factors based on a model of non-ideal characteristics of target memory to provide a training set of model parameters. An inference engine using the target memory technology to store the model parameters can have more stable results across a large number of engines.

In some implementations, the present disclosure relates to a method. The method includes obtaining a set of weights for a neural network comprising a plurality of nodes and a plurality of connections between the plurality of nodes. The method also includes identifying a first subset of weights and a second subset of weights based on the set of weights. The first subset of weights comprises weights that used by the neural network. The second subset of weights comprises weights that are prunable. The method further includes storing the first subset of weights in a first portion of a memory. A first error correction code is used for the first portion of the memory. The method further includes storing the second subset of weights in a second portion of the memory. A second error correction code is used for the second portion of the memory. The second error correction code is weaker than the first error correction code.

A method for generating a federated learning model is provided. The method includes obtaining images; obtaining sorting results of the images; and generating a trained federated learning model by training a federated learning model to be trained according to the images and the sorting results. The federated learning model to be trained is obtained after pruning a federated learning model to be pruned, and a pruning rate of a convolution layer in the federated learning model to be pruned is automatically adjusted according to a model accuracy during the pruning.

Non-intrusive, low-cost systems and methods allow designers to reduce headroom and safety margin requirements in the context of compute circuits, such as machine learning circuits, without increasing footprint or having to sacrifice computing capacity and other valuable resources. Various embodiments accomplish this by taking advantage of certain properties of machine learning circuits and using a CNN as a diagnostic tool for evaluating circuit behavior and adjusting circuit parameters to fully exploit available computing resources.