Other

Broadcom Corporation

Nvidia Corporation

Microsoft Technology Licensing, LLC

Avago Technologies General IP (Singapore) Pte. Ltd.

Siemens Corporation

Hao Fang Wu

Hooman Darabi

David Patrick Murphy

Hong-Jiang Zhang

Jonah Matthew Alben

Mingjing Li

Paulius Micikevicius

Hao Xu

Tong Fang

Mohyee Mikhemar

Ming Yiu Siu

Yunqiang Chen

Ujval J Kapasi

Steven Jay Randazzo

Michael Todd Gillam

Dilip Sequeira

Yilin Zhang

Alex Fit-Florea

Uri Kartoun

Renato Reder Cazangi

Alexey P Kontsevoy

Slawomir Kierat

Szymon Migacz

Maxim Milakov

Mario F Solorio

Zacky Zhou

Ruogu Wang

In training a deep neural network using reduced precision, gradient computation operates on larger values without affecting the rest of the training procedure. One technique trains the deep neural network to develop loss, scales the loss, computes gradients at a reduced precision, and reduces the magnitude of the computed gradients to compensate for scaling of the loss. In one example non-limiting arrangement, the training forward pass scales a loss value by some factor S and the weight update reduces the weight gradient contribution by 1/S. Several techniques can be used for selecting scaling factor S and adjusting the weight update.

Loss-scaling for deep neural network training with reduced precision

Aspects of the present invention are directed to computer-implemented techniques for performing data compression and conversion between data formats of varying degrees of precision, and more particularly for improving the inferencing (application) of artificial neural networks using a reduced precision (e.g., INT8) data format. Embodiments of the present invention generate candidate conversions of data output, then employ a relative measure of quality to identify the candidate conversion with the greatest accuracy (i.e., least divergence from the original higher precision values). The representation can be then be used during inference to perform computations on the resulting output data.

Automated methods for conversions to a lower precision data format

A method, computer readable medium, and system are disclosed for rounding numerical values. A set of bits from an input value is identified as a rounding value. A second set of bits representing a second value is extracted from the input value and added with the rounding value to produce a sum. The sum is truncated to produce the rounded output value. Thus, the present invention provides a stochastic rounding technique that rounds up an input value as a function of a second value and a rounding value, both of which were obtained from the input value. When the second value and rounding value are obtained from consistent bit locations of the input value, the resulting output value is deterministic. Stochastic rounding, which is deterministic, is advantageously applicable in deep learning applications.

Stochastic rounding of numerical values

A receiver includes circuitry configured to determine one or more first local oscillator (LO) harmonics that correspond to one or more first spectrum segments of a down-converted received signal based on characteristics of the received signal. The one or more first LO harmonics of the received signal are amplified by applying one or more first transconductance coefficients to one or more first harmonic selective transinductance amplifiers (TIAs) corresponding to the one or more first spectrum segments. Digitized outputs of the plurality of harmonic selective TIAs are calibrated based on an amount of signal leakage between the plurality of spectrum segments of the down-converted received signal.

Harmonic selective full-band capture receiver with digital harmonic rejection calibration

A new resonance-based wireless power transfer system is disclosed. The new resonance-based wireless power transfer system includes a power transmitter and a power receiver. The power transmitter and the power receiver are both oscillator-based circuits, but unlike prior systems, their resonant frequencies are different. The difference between the transmitter's resonant frequency and the receiver's resonant frequency is set between a lower bound and an upper bound to ensure that the new resonance-based wireless power transfer system will have only one oscillation frequency while maintaining high power transfer efficiency over varied distances.

Wireless power transfer method and system

Recently proposed noise-cancelling receivers report a best case trade-off between noise figure and linearity for a matched wideband receiver. These receivers are further improved using a passive front-end gain. The front-end gain reduces the power requirements of the radio frequency transconductance stage, and potentially other stages where, e.g., smaller mixer switches may be employed.

Power reduction in noise-cancelling receivers

Because of associated disadvantages of narrow-band off-chip radio-frequency (RF) filtering, a mixer-first receiver front-end designed to tolerate blockers with reduced gain compression and noise factor degradation is disclosed. The mixer-first receiver front-end includes two separate down-conversion paths that help to reduce added noise and voltage gain prior to baseband filtering, which are critical factors in eliminating narrow-band off-chip RF filtering. The mixer-first receiver front-end can be used to support down-conversion of multiple different communication signals (e.g., cellular, WLAN, and WPAN communication) with different center frequencies. In addition, where it is not possible to use a single, mixer-first receiver front-end to down-convert two different communication signals with potentially different center frequencies due to the need for both communication signals to be down-converted simultaneously, two mixer-first receiver front-ends can be efficiently used by sharing an antenna via a common RF port.

Radio receiver co-existence

A radio receiver supporting cancellation of thermal and phase noise in a down-converted RF signal. An inbound RF signal and blocking signal are provided directly to a passive mixer for down-conversion into a first baseband signal having data, thermal noise, and reciprocal mixing (RM) noise components. The inbound signals are also provided to a transconductance circuit, the output of which is provided to a second passive mixer for conversion into a current signal having data and blocking signal components, and a RM image. The blocking signal component and the RM image are mixed with a second LO signal, derived from the blocking signal, to produce a RM noise cancellation signal. The data component of the current signal is converted into a second baseband signal having data and thermal noise components. The first baseband signal, second baseband signal and RM noise cancellation signal are then combined through harmonic recombination.

Highly linear receiver front-end with thermal and phase noise cancellation

The described concepts relate to automated patient disposition. One example can receive a clinician's disposition for a patient. This implementation can perform parameter-based cluster analysis on the patient and a set of patients to identify a sub-set of the patients with which the patient has a high similarity. This example can also cause a graphical user interface to be generated that conveys parameters from the sub-set of the patients and the patient.

Automated clustering for patient disposition

Certain exemplary embodiments comprise a method, which can comprise determining an image of a predetermined physiological structure of a patient. The image can be determined based upon a first set of image data of the predetermined physiological structure of the patient. The image can be based upon a second set of image data of the predetermined physiological structure of the patient. The image can be determined based upon an iteratively adjusted movement of the patient.

Device systems and methods for imaging

Methods and apparatuses are provided for detecting blur within digital images using Cepstrum analysis blur detection techniques that are able to detect motion blur and/or out-of-focus blur.

Hierarchical scheme for blur detection in a digital image

Methods and apparatuses are provided for detecting blur within digital images using wavelet transform and/or Cepstrum analysis blur detection techniques that are able to detect motion blur and/or out-of-focus blur.

Hierarchical scheme for blur detection in digital image using wavelet transform

A manually operated pallet changer system for use with automatic machine tools includes a pallet changer and a compatible machine tool mounted pallet receiver. The pallet changer includes a massive, floor supported base to which a cantilevered first platform is pivotally attached for rotation between a pallet-exchange position and a work-setup position. A second platform rests atop the first platform and supports a pair of work pallet guides. The two platforms are rotatably attached near their centers so that the pallet guides can be interchanged by rotating the upper platform about its center by 180&deg;. The configuration permits the platforms to be rotated as a unit away from a machine tool to the work-setup position. Work pallets are assembled and then the platforms are rotated together into the pallet-exchange position in which one of the pallet guides is aligned for transferring a work pallet to the machine tool mounted pallet receiver. The receiver includes an air bearing upper surface for simplifying the movement and positioning of a work pallet, a deployable clamping plate, and a pair of retractable locating pins for engaging openings on the underside of a work pallet to insure precision alignment during a machining operation.

Growing community of inventors

Chevy Chase, MD, United States of America

Hao Fang Wu