Abstract:
Certain aspects of the present disclosure provide techniques and apparatus for surrogated federated learning. A set of intermediate activations is received at a trusted server from a node device, where the node device generated the set of intermediate activations using a first set of layers of a neural network. One or more weights associated with a second set of layers of the neural network are refined using the set of intermediate activations, and one or more weight updates corresponding to the refined one or more weights are transmitted to a federated learning system.
Abstract:
Certain aspects of the present disclosure provide techniques and apparatus for performing mathematical operations on a processor. The method generally includes initializing at least a portion of weight data for a machine learning model in a first memory component associated with a processor. Input data is stored in a second memory component coupled with the processor. Operations using the machine learning model are executed, via a functional unit associated with the processor, based on the at least the portion of the weight data and the input data. A result of the operations using the machine learning model are stored in the second memory component.
Abstract:
Certain aspects of the present disclosure provide techniques for desparsified convolution. An activation tensor is received, and a convolution output is generated for the activation tensor, comprising: selecting a subset of weight elements, corresponding to a set of non-zero elements in the activation tensor, from a weight tensor, and multiplying the set of non-zero elements and the set of weight elements.
Abstract:
Methods, devices, and instructions for performing a reverse translation lookaside buffer (TLB) look-up using a physical address input, including obtaining with a first processor the physical address input, wherein the physical address input indicates a physical address corresponding to a shared memory, obtaining a first mask associated with a first virtual address from a first TLB entry within a TLB associated with the first processor, wherein the obtained first mask is a bit pattern, obtaining from the first TLB entry a first page frame number associated with the shared memory, applying the obtained first mask to the obtained first page frame number to generate a first value, applying the obtained first mask to the obtained physical address input to generate a second value, and comparing the first value and the second value to determine whether the first value and the second value match.
Abstract:
A user device having an active application periodically detects for a hands-off state of the user device. Upon detecting a hands-off state, the user device affects at least one of the active application and the user device. Prior to affecting at least one of the active application and the user device, the user device detects for one or more of a connection between the user device and another device, and an application of pressure to the user device. If either is detected the, the user device bypasses affecting at least one of the active application and the user device.
Abstract:
A PCD may include an active heat transfer system configured to transfer heat from the PCD to a docking device. The active heat transfer system may include a thermoelectric cooler, a heat pipe, or other heat transfer elements. The active heat transfer system may, based on temperature measurements, be activated when the PCD is coupled to the docking device.
Abstract:
A method, an apparatus, and a computer program product for thermal management are provided. The apparatus may be a mobile device, a UE, a base station, a tablet computer, a smart watch, a head-mounted display, a portable media player, a personal navigation device, a wearable device, etc. The apparatus determines a direction of a natural draft relative to the apparatus. The apparatus generates an airflow within the apparatus in the direction of the natural draft. To determine the direction of the natural draft, the apparatus may detect an orientation of the apparatus and determine an upward direction based on the detected orientation. The upward direction may be generally aligned with the natural draft.
Abstract:
Certain aspects of the present disclosure provide techniques for desparsified convolution. A weight tensor having unstructured sparsity is accessed, and a densified weight tensor is generated based on the weight tensor by directionally squeezing the weight tensor to remove sparse values, and generating a sparsity map based on the directional squeezing. The densified weight tensor and sparsity map are output for use in a convolutional neural network.
Abstract:
Certain aspects of the present disclosure provide techniques for efficient depthwise convolution. A convolution is performed with a compute-in-memory (CIM) array to generate CIM output, and at least a portion of the CIM output corresponding to a first output data channel, of a plurality of output data channels in the CIM output, is written to a digital multiply-accumulate (DMAC) activation buffer. A patch of the CIM output is read from the DMAC activation buffer, and weight data is read from a DMAC weight buffer. Multiply-accumulate (MAC) operations are performed with the patch of CIM output and the weight data to generate a DMAC output.
Abstract:
A method, an apparatus, and a computer program product for managing a display device are provided. The apparatus may be a wearable device. The apparatus acquires data associated with a foreground corresponding to an image displayed on a display of the wearable device. The apparatus determines an image characteristic of the foreground based on the acquired data. The apparatus detects a change in the image displayed on the display. The apparatus adjusts, upon the detection of the change in the image, a display characteristic of the image based on the image characteristic of the foreground and the change in the image.