Qualcomm Incorporated

Jin Won Lee

Shankar Sadasivam

Li Ding

Xinzhou Wu

Rashid Ahmed Akbar Attar

Vidya Narayanan

Feng Han

George Cherian

Santosh Paul Abraham

Yang Yang

Edward Harrison Teague

Christopher G Lott

Jamie Menjay Lin

Ashwin Deepak Swaminathan

Lukas Daniel Kuhn

Yash Sanjay Bhalgat

Anusha Lalitha

Alexandros Manolakos

Sanjiv Nanda

Jilei Hou

Chirag Sureshbhai Patel

Kambiz Azarian Yazdi

Leonard Henry Grokop

Fatih Murat Porikli

Carlos Manuel Puig

Rexford Alan Hill

Disha Ahuja

Tijmen Pieter Frederik Blankevoort

Pawel Lukowicz

A method performed by a computing device includes determining a partition for depth-first processing by a multi-layer artificial neural network (ANN) of the computing device. The computing device comprising a processor, on-chip memory, and off-chip memory. The first partition determined based on an amount of on-chip memory used by the first partition, an available amount of on-chip memory, and a size of a write back to the off-chip memory. The method also includes processing, at the device via the multi-layer ANN, an input, using the depth-first processing in accordance with the partition.

Depth-first deep convolutional neural network inference

Certain aspects of the present disclosure provide techniques for improved machine learning using gradient pruning, comprising computing, using a first batch of training data, a first gradient tensor comprising a gradient for each parameter of a parameter tensor for a machine learning model; identifying a first subset of gradients in the first gradient tensor based on a first gradient criteria; and updating a first subset of parameters in the parameter tensor based on the first subset of gradients in the first gradient tensor.

Gradient pruning for efficient training of machine learning models

Various embodiments include methods and devices for implementing automatic grammar augmentation for improving voice command recognition accuracy in systems with a small footprint acoustic model. Alternative expressions that may capture acoustic model decoding variations may be added to a grammar set. An acoustic model-specific statistical pronunciation dictionary may be derived by running the acoustic model through a large general speech dataset and constructing a command-specific candidate set containing potential grammar expressions. Greedy based and cross-entropy-method (CEM) based algorithms may be utilized to search the candidate set for augmentations with improved recognition accuracy.

Automatic grammar augmentation for robust voice command recognition

A method for pruning weights of an artificial neural network based on a learned threshold includes determining a pruning threshold for pruning a first set of pre-trained weights of multiple pre-trained weights based on a function of a classification loss and a regularization loss. Weights are pruned from the first set of pre-trained weights when a first value of the weight is less than the pruning threshold. A second set of pre-trained weights of the multiple pre-trained weights is fine-tuned or adjusted in response to a second value of each pre-trained weight in the second set of pre-trained weights being greater than the pruning threshold.

Learned threshold pruning for deep neural networks

Certain aspects of the present disclosure provide techniques for performing piecewise pointwise convolution, comprising: performing a first piecewise pointwise convolution on a first subset of data received via a first branch input at a piecewise pointwise convolution layer of a convolutional neural network (CNN) model; performing a second piecewise pointwise convolution on a second subset of data received via a second branch input at the piecewise pointwise convolution layer; determining a piecewise pointwise convolution output by summing a result of the first piecewise pointwise convolution and a result of the second piecewise pointwise convolution; and providing the piecewise pointwise convolution output to a second layer of the CNN model.

Efficient inferencing with piecewise pointwise convolution

Some disclosed systems may include a microphone system having two or more microphones, an interface system and a control system. In some examples, the control system may be capable of receiving, via the interface system, audio data from two or more microphones of the microphone system, of determining a gesture location based, at least in part, on the audio data and of controlling one or more settings of the system based on the gesture location.

Audio-based device control

A method of operating a shared resource in a mobile device includes extracting a set of features from a plurality of subsystems of the mobile device. The set of features may be extracted from each subsystem of the plurality of subsystems requesting services from one or more shared resources of the mobile device. One or more parameter of the shared resource(s) may be determined based on the extracted set of features from the plurality of subsystems. The shared resource(s) may be operated based on the determined parameter(s).

Power state control of a mobile device

Aspects of the subject matter described in this disclosure can be implemented in a fall detection device and method. One or more motion sensors can access a user's acceleration data. The acceleration data can be segmented using a segmentation algorithm to identify a potential fall event. The segmentation algorithm can determine a cumulative sum of the acceleration data, where the cumulative sum is based on acceleration values being greater than or less than an acceleration threshold value, and a potential fall event can be identified where the cumulative sum is greater than a cumulative sum threshold value. Statistical features can be extracted from the segmented acceleration data and aggregated, and a determination can be made as to whether the potential fall event is a fall event based at least in part on the statistical features.

Devices and methods for fall detection based on phase segmentation

A system and method are disclosed that may selectively switch a station's association from a first AP to a second AP based, at least in part, on whether an indication of first RSSI values for the first AP is greater than an indication of second RSSI values for the second AP by more than a difference value. For some embodiments, the difference value may be dynamically adjusted based, at least in part, on the first RSSI values, on a connection history of the station, and/or on a user selection.

Scan channel reduction to improve Wi-Fi system power

System and methods are disclosed to use information available on the state of mobile devices in a heuristics-based approach to improve motion state detection. In one or more embodiments, information on the WiFi connectivity of mobile devices may be used to improve the detection of the in-transit state. The WiFi connectivity information may be used with sensor signal such as accelerometer signals in a motion classifier to reduce the false positives of the in-transit state. In one or more embodiments, information that a mobile device is connected to a WiFi access point (AP) may be used as heuristics to reduce the probability of falsely classifying the mobile device in the in-transit state when mobile device is actually in the hand of a relatively stationary user. Information on the battery charging state or the wireless connectivity of the mobile devices may also be used to improve the detection of in-transit state.

In-transit detection using low complexity algorithm fusion and phone state heuristics

Systems and methods share context information on a neighbor aware network. In one aspect, a context providing device receives a plurality of responses to a discovery query from a context consuming device, and tailors services it offers to the context consuming device based on the responses. In another aspect, a context providing device indicates in its response to a discovery query which services or local context information it can provide to the context consuming device, and also a cost associated with providing the service or the local context information. In some aspects, the cost is in units of monetary currency. In other aspects, the cost is in units of user interface display made available to an entity associated with the context providing device in exchange for the services or local context information offered to the context consuming device.

Systems and methods for sharing context information in a neighbor aware network

Methods, systems, computer-readable media, and apparatuses for inferring context are provided. In one potential implementation, first context information associated with a first duration is identified, second context information is accessed to determine a context segmentation boundary; and the first context information and the second context information is then aggregated to generate an inferred segmented aggregated context. In a further implementation, the first context information is used to average inferred contexts, and the context segmentation boundary is used to reset a start time for averaging the first context information.

Aggregate context inferences using multiple context streams

Systems and methods share context information on a neighbor aware network. A method for communicating data in a wireless communications network is disclosed. The method includes receiving, by a device, a first message from a station, decoding the message to determine service information, the service information identifying a service provided by the station, generating a second message, wherein the second message is generated to indicate the service provided by the station and service information of the device, and transmitting, by the device, the second message to a remote station.

Growing community of inventors

San Diego, CA, United States of America

Jin Won Lee