International Business Machines Corporation

Jui-Hsin Lai

Li-Wen Hung

Yinglong Xia

Chia-Yu Chen

Ching-Yung Lin

Ko-Tao Lee

Lifeng Nai

Toyotaro Suzumura

Yu Ma

Conglei Shi

Blaine H Dolph

David M Lubensky

Pei-Yun S Hsueh

Pierce I-Jen Chuang

Ruichi Yu

Chun-Fu Chen

Wan-Yi Lin

Guangnan Ye

Toyotaro Suzumara

Lifeng Nail

A Static Random Access Memory (SRAM) device in a binary neural network is provided. The SRAM device includes an SRAM inference engine having an SRAM computation architecture with a forward path that include multiple SRAM cells forming a chain of SRAM cells such that an output of a given one of the multiple SRAM cells is an input to a following one of the multiple SRAM cells. The SRAM computation architecture is configured to compute a prediction from an input. The SRAM computation architecture is configured to store ternary data and perform local computations on the ternary data.

Inference focus for offline training of SRAM inference engine in binary neural network

A Static Random Access Memory (SRAM) device in a binary neural network is provided. The SRAM device includes an SRAM inference engine having an SRAM computation architecture with a forward path that include multiple SRAM cells. The multiple SRAM cells are configured to form a chain of SRAM cells such that an output of a given one of the multiple SRAM cells is an input to a following one of the multiple SRAM cells. The SRAM computation architecture is configured to compute a prediction from an input.

A query image is obtained. In a database including a plurality of reference image graphs, at least one of the reference image graphs, with feature vectors similar to the query image, is identified. Image querying is carried out by graph traversal on the at least one of the reference image graphs with the feature vectors similar to the query image. An image from the at least one of the reference image graphs having a highest matching score in the graph traversal is returned as a response to the query image. techniques for building the database are also disclosed.

Graph-based online image queries

Method and apparatus for optimizing a convolutional neural network (CNN). A respective measure of importance is calculated for each of a plurality of elements within a CNN. A first one of the measures of importance is calculated by back propagating a second one of the measures of importance through the CNN. One or more of the plurality of elements is pruned from the CNN, based on the calculated measures of importance.

Pruning redundant neurons and kernels of deep convolutional neural networks

Methods and systems for activity monitoring include capturing an infrared image of an environment that comprises at least one patient being monitored and at least one infrared-emitting tag. A relationship between the patient being monitored and the at least one infrared-emitting tag is determined. An activity conducted by the patient being monitored is determined based on the relationship between the patient being monitored and the at least one infrared-emitting tag. A course of treatment for the patient being monitored is adjusted based on the determined activity.

Infrared detectors and thermal tags for real-time activity monitoring

Method and apparatus for monitoring eating behavior are disclosed. A wearable device including an imaging sensor, an electromyography (EMG) sensor, and a processing unit is configured to collect EMG data via the EMG sensor, transmit the EMG data to a computing device, receive a control signal to capture one or more images via the imaging sensor, capture, one or more images, and transmit the one or more images. A computing device is disclosed including a processor and a memory containing a machine learning model and computer program code that, when executed, performs an operation. The operation includes receiving EMG data, processing the EMG data as input to a trained machine learning model, transmitting the control signal to capture one or more images via the imaging sensor, receiving the one or more images, and determining nutritional content of the food or beverage using the one or more images.

System for monitoring eating habit using a wearable device

Methods of fabricating a probe are described. In an example, a structure may be formed on a surface of a substrate. The structure may include the probe, a hinge, and an anchor arranged linearly, where an angle is formed between the probe and the hinge. The hinge may be positioned between the probe and the anchor, and the structure may be parallel to the substrate. An amount of solder may be deposited on an area of the structure that spans from a portion of the probe to a portion of the anchor, and across the hinge. The deposited solder may be reshaped by an execution of a solder reflow process. The reshape of the deposited solder may cause the probe to rotate about the hinge in order to reduce the angle between the probe and the hinge.

Planar fabrication of micro-needles

A Resilient Distributed Dataset (RDD) processing method, system, and computer program product include sending a different portion including a global identifier and different values of a message from a receiver to a sender and at a receiver side, an already-processed message is extracted from an RDD object repository corresponding to the global identifier of the different portion and the incoming differential portion is applied to the already-processed message.

Differential processing mechanism for spark-based graph computing background

A thermal tag for activity monitoring. The thermal tag includes a base layer having a plurality of metal lines to provide a conductive path, and a pattern layer having one or more infrared emitting features positioned over portions of the conductive path, wherein at least one infrared emitting feature couples to the conductive path to emit a predetermined infrared pattern in accordance with nearby activity.

Thermal tags for real-time activity monitoring and methods for fabricating the same

A remote control system detects a directional signal at an alerting sensor package; activates an appliance to accept a control signal in response to the directional signal; accepts the control signal at the appliance from a muscle sensor package; and adjusts operation of the appliance in response to the control signal.

Remote control with muscle sensor and alerting sensor

A method includes analyzing a preview image made by an image sensor when operating in a preview image mode to determine if the preview image contains some portion that corresponds to information that is considered not subject to being imaged. The method further includes, if it is determined that the preview image does comprise some image portion that corresponds to information that is considered not subject to being imaged, disabling the image sensor from capturing an image that would correspond to the preview image. Analyzing can include accessing a data storage that contains representations of information that is considered not subject to being imaged for comparison with image features extracted from the preview image, where the data storage is at least one of contained in a device that also contains the image sensor, or is located remotely from the device and can be accessed via a wireless connection.

Method, system and computer program product for selective image capture

A random number signal generator used for performing dropout or weight initialization for a node in a neural network. The random number signal generator includes a transistor which generates a random noise signal. The transistor includes a substrate, source and drain regions formed in the substrate, a first insulating layer formed over a channel of the transistor, a first trapping layer formed over the first insulating layer, a second insulating layer formed over the first trapping layer, and a second trapping layer formed over the second insulating layer. One or more traps in the first and second trapping layers are configured to capture or release one or more carriers flowing through the channel region. The random noise signal is generated as a function of one or more carrier being captured or released by the one or more traps.

True random generator (TRNG) in ML accelerators for NN dropout and initialization

A computer-implemented method, computer program product, and system for determination of critical parts and component correlations in a circuit using a correlation graph and centrality analysis including; receiving a circuit layout portion of a larger circuit layout, converting the circuit layout portion into a correlation graph representing components as nodes and connecting wires as edges, determining, using ground truth and Na&#xef;ve Bayes to determine correlation weighting, scaling the correlation graph to represent the larger circuit, and presenting the larger correlation graph on a graphical user interface (GUI).

Growing community of inventors

White Plains, NY, United States of America

Jui-Hsin Lai