A system for parallel photonic computation, preferably including one or more source modules, a plurality of multiplication modules, and a plurality of summation modules. In one embodiment, each multiplication module can include a set of input modulators, a splitter, and a plurality of multiplication banks. Each summation module can include one or more detectors. Each summation module preferably receives an output from multiple multiplication modules and computes the sum of all channels of all the received outputs. A method for parallel photonic computation, preferably including generating input signals, computing products, and computing sums.

This application provides a method, a terminal-side device, and a cloud-side device for data processing and a terminal-cloud collaboration system. The method includes: sending, by the terminal-side device, a request message to the cloud-side device; receiving, by the terminal-side device, a second neural network model that is obtained by compressing a first neural network model and that is sent by the cloud-side device, where the first neural network model is a neural network model on the cloud-side device that is used to process the cognitive computing task, and a hardware resource required when the second neural network model runs on the terminal-side device is within an available hardware resource capability range of the terminal-side device; and processing, by the terminal-side device, the cognitive computing task based on the second neural network model.

A camera simulation system or method or process for simulating performance of a camera for a vehicle includes providing a camera having a lens and imager and providing a learning algorithm. Image data is captured from a raw image input of the camera and the captured image data and raw image input are output from the camera. The output image data and the raw image input are provided to the learning algorithm. The learning algorithm is trained to simulate performance of the lens and/or the imager using the output captured image data and the raw image data input. The performance of the lens and/or the imager is simulated responsive to the learning algorithm receiving raw images.

Accelerating photonic and opto-electronic technologies requires breaking current limits of modern chip-scale photonic devices. While electronics and computer technologies have benefited from Moore's Law scaling, photonic technologies are conventionally limited in scale by the wavelength of light. Recent sub-wavelength optical devices use nanostructures and plasmonic devices but still face fundamental performance limitations arising from metal-induced optical losses and resonance-induced narrow optical bandwidths. The present disclosure instead confines and guides light at deeply sub-wavelength dimensions while preserving low-loss and broadband operation. The wave nature of light is used while employing metal-free (all-dielectric) nanostructure geometries which effectively pinch light into ultra-small active volumes, for potentially about 100-1000 reduction in energy consumption of active photonic components such as phase-shifters. The present disclosure could make possible all-optical and quantum computing devices which require extreme optical confinement to achieve efficient light-matter interactions.

A computer-implemented method includes receiving a neural network model that includes a tensor operation, dividing the tensor operation into a set of sub-operations, and generating instructions for performing a plurality of sub-operations of the set of sub-operations on respective computing engines of a plurality of computing engines on a same integrated circuit device or on different integrated circuit devices. Each sub-operation of the set of sub-operations generates a portion of a final output of the tensor operation. An inference is made based on a result of a sub-operation of the plurality of sub-operations, or based on results of the plurality of sub-operations.

A device for image classification comprising a convolutional neural network configured to generate a plurality of probability values, each probability value being linked to a respective one of a plurality of predetermined classes and indicating the probability that the image or a pixel of the image is associated with the respective class, and the convolutional neural network comprises a plurality of convolutional blocks and each of the convolutional blocks comprises: a first convolutional layer configured to perform a pointwise convolution using a first kernel, a second convolutional layer configured to perform a depthwise convolution using a second kernel, wherein the second kernel has one of a single row and a single column, a third convolutional layer configured to perform a depthwise convolution using a third kernel, wherein the third kernel has a single column if the second kernel has a single row, and the third kernel has a single row if the second kernel has a single column, and a fourth convolutional layer configured to perform a convolution using a fourth kernel.

In some examples, a system for training a neural network can include a processor to detect a trained neural network application. The processor can also detect a set of images, wherein the neural network application is not trained with the set of images. Additionally, the processor can train an adapter network based on the trained neural network application and the set of images, wherein the adapter network is to be trained by freezing weights of the trained neural network and modifying weights of the adapter network. Furthermore, the processor can use the trained adapter network to process at least one additional image, the processed additional image to be transmitted to the trained neural network to generate an output signal.

Methods and apparatus for training a matrix-based differentiable program using a photonics-based processor. The matrix-based differentiable program includes at least one matrix-valued variable associated with a matrix of values in a Euclidean vector space. The method comprises configuring components of the photonics-based processor to represent the matrix of values as an angular representation, processing, using the components of the photonics-based processor, training data to compute an error vector, determining in parallel, at least some gradients of parameters of the angular representation, wherein the determining is based on the error vector and a current input training vector, and updating the matrix of values by updating the angular representation based on the determined gradients.

Methods and apparatus for training a matrix-based differentiable program using a photonics-based processor. The matrix-based differentiable program includes at least one matrix-valued variable associated with a matrix of values in a Euclidean vector space. The method comprises configuring components of the photonics-based processor to represent the matrix of values as an angular representation, processing, using the components of the photonics-based processor, training data to compute an error vector, determining in parallel, at least some gradients of parameters of the angular representation, wherein the determining is based on the error vector and a current input training vector, and updating the matrix of values by updating the angular representation based on the determined gradients.

A system comprising: a processor; a data bus coupled to the processor; and a non-transitory, computer-readable storage medium embodying computer program code, the non-transitory, computer-readable storage medium being coupled to the data bus. The computer program code interacting with a plurality of computer operations and comprising instructions executable by the processor and configured for: receiving data from a data source; processing the data, the processing comprising performing a parsing operation on the data, the processing the data identifying a plurality of knowledge elements based upon the parsing operation, the parsing operation comprising ranking of parse options; and, storing the knowledge elements within the cognitive graph as a collection of knowledge elements, the storing universally representing knowledge obtained from the data.