Intel Corporation

Samsung Electronics Co., Ltd.

Indian Institute of Science

Seagate Technology International

Sandcraft, Inc.

Vinod Sharma

Hyung Jai Lee

Edward Thomas Grochowski

Monica Lucia Martinez-Canales

Malini Krishnan Bhandaru

Bharat Bhushan

Joseph Chang

John H Crawford

Dong Jun Lee

Haesung Kwon

Sudhir Kumar Singh

Neelesh B Mehta

Rita Chattopadhyay

Frank Ivan Morris

Debasis Baral

Bipin V Gami

Ron Kirkish

Gaurav Bansal

Wei Zhou

Na Young Kim

Yi Chen

Ralph Michael Kling

Young-Hoon Kim

Brian D Strom

Vivek Tiwari

Suri B Medapati

Peggy J Irelan

Vivek Joshi

Andrei Khurshudov

Sivakumar Makineni

Sajal Mittal

Soila Pertet Kavulya

Hareharan Nagarajan

Abdur Rakheeb

Kai Zheng

Varsha Parthasarathy

Nofil Fawad

Gregory S Matthews

Nandish Uppal Raravi

Sung Chang Lee

Uma Maheswara Rao Gangumalla

Niraj Rai

Rakesh Radhakrishnan Potty

Timothy Isaac Anderson

Qiyuan Gong

A quaternion deep neural network (QTDNN) includes a plurality of modular hidden layers, each comprising a set of QT computation sublayers, including a quaternion (QT) general matrix multiplication sublayer, a QT non-linear activations sublayer, and a QT sampling sublayer arranged along a forward signal propagation path. Each QT computation sublayer of the set has a plurality of QT computation engines. In each modular hidden layer, a steering sublayer precedes each of the QT computation sublayers along the forward signal propagation path. The steering sublayer directs a forward-propagating quaternion-valued signal to a selected at least one QT computation engine of a next QT computation subsequent sublayer.

Computationally-efficient quaternion-based machine-learning system

A machine-learning system includes a quaternion (QT) computation engine. Input data to the QT computation engine includes quaternion values, each comprising a real component and three imaginary components, represented as a set of real-valued tensors. A single quaternion value is represented as a 1-dimensional real-valued tensor having four real-valued components, wherein a first real-valued component represents the real component of the single quaternion value, and wherein a second, a third, and a fourth real-valued component each respectively represents one of the imaginary components. A quaternion-valued vector having a size N is represented as a 2-dimensional real-valued tensor comprising N 1-dimensional real-valued tensors. A quaternion-valued matrix having N&#xd7;M dimensions is represented as a 3-dimensional real-valued tensor comprising M 2-dimensional real-valued tensors comprising N 1-dimensional real-valued tensors.

Tensor-based computing system for quaternion operations

Systems and techniques for neural network training are described herein, a training set may be received for a neural network. Here, the neural network may comprise a set of nodes arranged in layers and a set of inter-node weights between nodes in the set of nodes. The neural network may then be iteratively trained to create a trained neural network. An iteration of the training may include generating a random unit vector and creating an update vector by calculating a magnitude for the random unit vector based on a degree that the random unit vector matches a gradient&#x2014;where the gradient is represented by a dual number. The iteration may continue by updating a parameter vector for an inter-node weight by subtracting the update vector from a previous parameter vector of the inter-node weight. The trained neural network may then be used to classify data.

Neural network training using generated random unit vector

A deep neural network (DNN) includes hidden layers arranged along a forward propagation path between an input layer and an output layer. The input layer accepts training data comprising quaternion values, outputs a quaternion-valued signal along the forward path to at least one of the hidden layers. At least some of the hidden layers include quaternion layers to execute consistent quaternion (QT) forward operations based on one or more variable parameters. A loss function engine produces a loss function representing an error between the DNN result and an expected result. QT backpropagation-based training operations include computing layer-wise QT partial derivatives, consistent with an orthogonal basis of quaternion space, of the loss function with respect to a QT conjugate of the one or more variable parameters and of respective inputs to the quaternion layers.

Gradient-based training engine for quaternion-based machine-learning systems

Examples include techniques for backing up a file to long term 'cold' storage by using circuitry, and logic for execution by the circuitry, to receive a request to back up the file in a distributed file system to cold storage, to copy the file from at least one data node of the distributed file system to cold storage, to set a location of the file in cold storage in a name node of the distributed file system, and to set a length of the file to zero in the name node.

Method of efficient backup of distributed file system files with transparent data access

Methods and apparatus for implementing a current-mirror based level shifter circuit are provided. The current-mirror based level shifter circuit includes a current-mirror circuit, a feedback control circuit, a power down circuit and a plurality of inverter circuits. The apparatus is configured to provide a wide voltage shifting range using the current-mirror based level shifter circuit. The apparatus comprising a feedback loop with two diode connected transistors may provide a constant drivability to the node that drives the output, when a current-mirror circuit is turned-off.

Current-mirror based level shifter circuit and methods for implementing the same

Various systems and methods for implementing context-based digital signal processing are described herein. An object detection system includes a processor to: access sensor data from a first sensor and a second sensor integrated in a vehicle; access an operating context of the vehicle; assign a first weight to a first object detection result from sensor data of the first sensor, the first weight adjusted based on the operating context; assign a second weight to a second object detection result from sensor data of the second sensor, the second weight adjusted based on the operating context; and perform a combined object detection technique by combining the first object detection result weighted by the first weight and the second object detection result weighted by the second weight.

Context-based digital signal processing

Machine vision processing includes capturing 3D spatial data representing a field of view and including ranging measurements to various points within the field of view, applying a segmentation algorithm to the 3D spatial data to produce a segmentation assessment indicating a presence of individual objects within the field of view, wherein the segmentation algorithm is based on at least one adjustable parameter, and adjusting a value of the at least one adjustable parameter based on the ranging measurements. The segmentation assessment is based on application of the segmentation algorithm to the 3D spatial data, with different values of the at least one adjustable parameter value corresponding to different values of the ranging measurements of the various points within the field of view.

Adaptive processing of spatial imaging data

A system using cooperative communication with rateless codes is presented that uses communication transmission aspects of cooperative communication with rateless codes over Rayleigh fading channels and queuing aspects for buffering messages at intermediate relays. The system transmits a subsequent message while a current message is en route to the destination by receiving and buffering the current message in queues at intermediate relays. A relay with a best instantaneous communication link to the source may receive a message first and forward the received message to the destination. Alternatively, if inter-relay communication links are strong, all relays may cooperate simultaneously.

Queued cooperative wireless networks configuration using rateless codes

Disclosed is a head gimbal assembly for a hard disk drive that includes a head pivotally connected to a flexure. The flexure has a longitudinal axis. The assembly includes first and second micro-actuators that are located between the flexure and the head and spaced from the longitudinal axis. By way of example, the first and second micro-actuators may be thin piezo-electric transducers. Control signals can be provided to the micro-actuators to vary the pitch and/or roll of the head.

In-situ dynamic pitch and roll adjustment in hard disk drives

The head gimbal assembly includes a slider and a micro-actuator assembly sharing power for the micro-actuator assembly to aid in positioning the slider and for an amplifier included in the slider to generate an amplified read signal when the slider read accesses data on a rotating disk surface included in a hard disk drive. The slider includes a read-write head providing a read differential signal pair to an amplifier to generate an amplified read signal reported when read accessing a rotating disk surface near the slider, which includes a read head employing a spin valve or employing a tunneling valve. Hard disk drive including a head stack assembly, which includes at least one of the head gimbal assemblies. Manufacturing the head gimbal assembly, the head stack assembly, and the hard disk drive, as well as these items as products of the invention's manufacturing processes.

Apparatus and method for head gimbal assembly sharing power to slider amplifier and micro-actuator in a hard disk drive

A magnetic head follows a track in a hard disk drive as positioned by micro-actuator and voice coil motor. Embodiments operate two control paths. The micro-actuator control path generates a version of micro-actuator control signal stimulating the micro-actuator. The voice coil motor control path generates a version of voice coil control signal, amplified based upon tuning gain to stimulate voice coil motor. A decoupling feedback filter decouples these control paths, using the micro-actuator control signal version. Either control path may include a notch filter. A track following command, with PES removed, directs the micro-actuator control. The servo-controller may digitally support the invention, which may include the servo-controller program system. An implementation may be optimized within a hard disk drive, possibly as part of the manufacturing process. The hard disk drive is a product of that process.

Method and apparatus of dual stage servo control with dual control paths and decoupling feedback for track following in a hard disk drive

Method of estimating the stroke sensitivity of micro-actuator coupled with slider and its read-write head. Micro-actuator stimulus signal used to drive micro-actuator, inducing noise into lateral positioning of read-write head near track by voice coil motor to create PES. Lateral position noise derived from Position Error Signal. Stroke sensitivity estimated from lateral position noise and micro-actuator stimulus signal. Apparatus supporting method estimating stroke sensitivity: servo controller, embedded circuit including servo-controller, and hard disk drive may include servo-controller and/or embedded circuit. Method making servo-controller, embedded circuit, and/or hard disk drive. Servo-controller, embedded circuit, and hard disk drive are products of these processes.

Method and apparatus for estimating micro-actuator stroke sensitivity in a hard disk drive

A slider, flexure finger, head suspension assembly, head gimbal assembly, main flex circuit, actuator assembly, spindle motor, embedded circuit, and hard disk drive using and/or supporting method of estimating flying height of read-write head near rotating disk surface. Method senses leakage current between slider and spindle motor to create potential difference measurement, and converts potential difference measurement to create estimate of flying height. Method controlling flying height using method for estimating by adjusting vertical actuation signal to micro-actuator assembly. The invention includes methods of manufacturing and their products.

Method and apparatus for sensing leakage current to estimate and possibly adjust flying height of a read-write head in a hard disk drive

Growing community of inventors

Los Gatos, CA, United States of America

Vinod Sharma