Abstract: A computer-implemented method of training a streaming speech recognition model that includes receiving, as input to the streaming speech recognition model, a sequence of acoustic frames. The streaming speech recognition model is configured to learn an alignment probability between the sequence of acoustic frames and an output sequence of vocabulary tokens. The vocabulary tokens include a plurality of label tokens and a blank token. At each output step, the method includes determining a first probability of emitting one of the label tokens and determining a second probability of emitting the blank token. The method also includes generating the alignment probability at a sequence level based on the first probability and the second probability. The method also includes applying a tuning parameter to the alignment probability at the sequence level to maximize the first probability of emitting one of the label tokens.

Abstract: Two-pass automatic speech recognition (ASR) models can be used to perform streaming on-device ASR to generate a text representation of an utterance captured in audio data. Various implementations include a first-pass portion of the ASR model used to generate streaming candidate recognition(s) of an utterance captured in audio data. For example, the first-pass portion can include a recurrent neural network transformer (RNN-T) decoder. Various implementations include a second-pass portion of the ASR model used to revise the streaming candidate recognition(s) of the utterance and generate a text representation of the utterance. For example, the second-pass portion can include a listen attend spell (LAS) decoder. Various implementations include a shared encoder shared between the RNN-T decoder and the LAS decoder.

Abstract: An image-based fluid characterization system may include a container configured to fully enclose a fluid, one or more illumination sources configured illuminate an interior portion of the container with illumination and one or more imaging systems. Each of the one or more imaging systems may include a multi-pixel sensor and an optical relay, where the multi-pixel sensor generates one or more images of the fluid in response to the illumination via the optical relay. The system may further include a controller to generate one or more measurements of the fluid based on the one or more images.

Abstract: An adaptive system and a method for adjusting a parameter in the adaptive system includes operating an adaptive system with an output signal produced from an input signal applied to an input, in which a parameter with a finite range is determined based on a difference between the output signal and a target output signal. In one example the parameter with the finite range is a fixed-point parameter or an analog parameter. The parameter is accessed from the adaptive system. A probability distribution of the parameter is assigned. The finite range for the parameter is updated based on the probability distribution which has been assigned. The probability distribution function may be updated alone with the finite range of the parameter. The probability distribution may be derived from one or more historical values of the parameter, and a plurality of system parameters belonging to an identical category of data as the parameter.

Abstract: Photonic units for vector-vector multiplication, matrix-vector multiplication, matrix-matrix multiplication, batch matrix-matrix multiplication, and tensor-tensor multiplication are described. Multiplications are through coherent mixing and square-law detection. There are many dimensions—wavelength, vector mode, quadrature, and three dimensions of space—that can be used to construct photonic accelerators. The encoded input vector or input matrix is fanned out into a desired number of copies and mixed with the corresponding encoded local oscillators containing the weight vectors comprising the weight matrix. Any subset of two (three) dimensions can be used to construct photonic accelerators for matrix-vector (matrix-matrix) multiplications. Multiple dimensions can be combined into a hyperdimension to increase the scalability.

Abstract: A flexible, artifact-free, and lensless fiber-based imaging system for biological objects. This system combines image reconstruction by a trained deep neural network with low-loss image transmission through disordered glass-air Anderson localized optical fiber. High quality images of biological objects can be obtained using short (few centimeters) or long (more than one meter) segments of disordered fiber with and without fiber bending. The deep neural network can also be designed to perform image classification. The system provides the unique property that the training performed within a straight fiber setup can be utilized for high fidelity reconstruction/classification of images that are transported through either straight or bent fiber making retraining for different bending situations unnecessary.

Abstract: An adaptive system and a method for adjusting a parameter in the adaptive system includes operating an adaptive system with an output signal produced from an input signal applied to an input, in which a parameter with a finite range is determined based on a difference between the output signal and a target output signal. In one example the parameter with the finite range is a fixed-point parameter or an analog parameter. The parameter is accessed from the adaptive system. A probability distribution of the parameter is assigned. The finite range for the parameter is updated based on the probability distribution which has been assigned. The probability distribution function may be updated alone with the finite range of the parameter. The probability distribution may be derived from one or more historical values of the parameter, and a plurality of system parameters belonging to an identical category of data as the parameter.

Abstract: A flexible, artifact-free, and lensless fiber-based imaging system for biological objects. This system combines image reconstruction by a trained deep neural network with low-loss image transmission through disordered glass-air Anderson localized optical fiber. High quality images of biological objects can be obtained using short (few centimeters) or long (more than one meter) segments of disordered fiber with and without fiber bending. The deep neural network can also be designed to perform image classification. The system provides the unique property that the training performed within a straight fiber setup can be utilized for high fidelity reconstruction/classification of images that are transported through either straight or bent fiber making retraining for different bending situations unnecessary.

Abstract: Novel and advantageous systems and methods for performing X-ray imaging by extracting X-ray phase-shift and/or dark-field information through a detector that has built-in G2 functionality are provided. Grating translation can be replaced by an electrical operation in the detection procedure, thereby eliminating the need for the analyzer grating and the typical mechanical stepping process.

Abstract: A kinetic energy conversion device of a punching machine includes a punching table, a power unit, an air compressor, a clutch, a gas storage tank, a pressure detector and a generator. The power unit, the air compressor, the clutch, the gas storage tank, the pressure detector, and the generator are all disposed on the punching table. The power unit drives the clutch, air compressor, and generator to generate electrical power and high pressure gas. Therefore, the air compressor can operate without using additional energy. Compared with conventional punching equipment and the air compressor externally disposed and connected to the punching equipment, a large amount of electricity cost can be saved, achieving a power saving effect.

Abstract: An imaging system operative for providing a volumetric molecular image of an object is disclosed. The imaging system interrogates the object with structured x-ray radiation while continuous relative motion between the object and source is induced during a measurement period. As the radiation passes through the object, the radiation scatters based on the molecular composition within the object, and the scattering changes as a function of time due to the relative motion between the source and object. Coherent scatter radiation is detected and processed to reconstruct an estimate of the three-dimensional molecular structure of the object using a reconstruction algorithm, such as maximum likelihood estimation.

Abstract: Novel and advantageous systems and methods for performing X-ray imaging by extracting X-ray phase-shift and/or dark-field information through a detector that has built-in G2 functionality are provided. Grating translation can be replaced by an electrical operation in the detection procedure, thereby eliminating the need for the analyzer grating and the typical mechanical stepping process.

Abstract: An e-Petri dish comprising a transparent layer having a specimen surface and a light detector configured to sample a sequence of sub-pixel shifted projection images of a specimen located on the specimen surface. The sub-pixel shifted projection images associated with light from a plurality of illumination angles provided by an illumination source.

Abstract: Methods of rapid distinction between growing cells and debris, which determine a time-lapse movie of specimen images, track features of each entity, and categorize each entity as growing cells or debris.

Abstract: An imaging system operative for providing a volumetric molecular image of an object is disclosed. The imaging system interrogates the object with structured x-ray radiation while continuous relative motion between the object and source is induced during a measurement period. As the radiation passes through the object, the radiation scatters based on the molecular composition within the object, and the scattering changes as a function of time due to the relative motion between the source and object. Coherent scatter radiation is detected and processed to reconstruct an estimate of the three-dimensional molecular structure of the object using a reconstruction algorithm, such as maximum likelihood estimation.

Abstract: A light guided pixel having a guide layer and a light detector layer. The guide layer has a light guide. The light detector layer has a light detecting element that receives light channeled by the light guide. The light guide may include a filter for channeling emissions to the light detecting element.

Abstract: Talbot imaging systems comprising a Talbot element, a phase gradient generating device, a light detector, and a processor. The Talbot element repeats a Talbot image at a distance from the Talbot element. The phase gradient generating device scans the Talbot image at a plane at the distance from the Talbot element by incrementally changing a phase gradient of a light field incident the Talbot element. As the Talbot image is scanned, the light detector captures time varying data associated with light altered by an object located at the distance from the Talbot element. The processor reconstructs an image of the object based on the time-varying light data.

Abstract: A Talbot-illuminated imaging system for focal plane tuning, the device comprising a Talbot element, a tunable illumination source, a scanning mechanism, a light detector, and a processor. The element generate san array of focused light spots at a focal plane. The tunable illumination source shifts the focal plane to a plane of interest by adjusting a wavelength of light incident the Talbot element. The scanning mechanism scans an object across an array of focused light spots in a scanning direction. The light detector determines time-varying light data associated with the array of focused light spots as the object scans across the array of light spots. The processor constructs an image of the object based on the time-varying data.

Abstract: Embodiments of the present invention relate to holographically illuminated imaging devices including a holographic element for transforming an illumination beam into a focal array of light spots, a scanning mechanism for moving an object across one or more light spots in the focal array of light spots, and a light detector for detecting light associated with the focal array of light spots and generating light data associated with the received light.

Abstract: Methods of rapid distinction between growing cells and debris, which determine a time-lapse movie of specimen images, track features of each entity, and categorize each entity as growing cells or debris.