NDLI: Zero and data reuse-aware fast convolution for deep neural networks on GPU

Please wait, while we are loading the content...

A practical methodology to validate the statistical behavior of bloom filters

Scalable and realistic benchmark synthesis for efficient NoC performance evaluation: a complex network analysis approach

An accurate and flexible early memory system power evaluation approach using a microcomponent method

Time in cyber-physical systems

A design to reduce write amplification in object-based NAND flash devices

How to enable software isolation and boost system performance with sub-block erase over 3D flash memory

Realizing erase-free SLC flash memory with rewritable programming design

IoT technologies for embedded computing: a survey

Distributed QoS management for internet of things under resource constraints

Evolving authentication design considerations for the internet of biometric things (IoBT)

Fast and cycle-accurate simulation of multi-threaded applications on SMP architectures using hybrid prototyping

Efficient design space exploration by knowledge transfer

Optimal functional-unit assignment and buffer placement for probabilistic pipelines

An overview of micron's automata processor

Enabling the high level synthesis of data analytics accelerators

Big data analytics on heterogeneous accelerator architectures

Going deeper than deep learning for massive data analytics under physical constraints

Fault injection at host-compiled level with static fault set reduction for SoC firmware robustness testing

Optimizing the location of ECC protection in network-on-chip

A disturbance-aware sub-block design to improve reliability of 3D MLC flash memory

Self-aware systems for the internet-of-things

Checkpoint aware hybrid cache architecture for NV processor in energy harvesting powered systems

Nano-engineered architectures for ultra-low power wireless body sensor nodes

Mitigation of homodyne crosstalk noise in silicon photonic NoC architectures with tunable decoupling

Making the internet-of-things a reality: from smart models, sensing and actuation to energy-efficient architectures

Web browser workload characterization for power management on HMP platforms

SPARTA: runtime task allocation for energy efficient heterogeneous many-cores

Energy-efficient mapping of real-time applications on heterogeneous MPSoCs using task replication

Industrial IoT lifecycle via digital twins

Security and privacy challenges in IoT-based machine-to-machine collaborative scenarios

Human-machine communication for assistive IoT technologies

AsyncStripe: I/O efficient asynchronous graph computing on a single server

Zero and data reuse-aware fast convolution for deep neural networks on GPU

Runtime configurable deep neural networks for energy-accuracy trade-off

Types of computational self-awareness and how we might implement them

On-the-fly computing: self-aware heterogeneous multi-cores

Conquering MPSoC complexity with principles of a self-aware information processing factory

Zero and data reuse-aware fast convolution for deep neural networks on GPU

Content Provider	ACM Digital Library
Author	Ahn, Junwhan Yoo, Sungjoo Park, Hyunsun Kim, Dongyoung
Abstract	Convolution operations dominate the total execution time of deep convolutional neural networks (CNNs). In this paper, we aim at enhancing the performance of the state-of-the-art convolution algorithm (called Winograd convolution) on the GPU. Our work is based on two observations: (1) CNNs often have abundant zero weights and (2) the performance benefit of Winograd convolution is limited mainly due to extra additions incurred during data transformation. In order to exploit abundant zero weights, we propose a low-overhead and efficient hardware mechanism that skips multiplications that will always give zero results regardless of input data (called ZeroSkip). In addition, to leverage the second observation, we present data reuse optimization for addition operations in Winograd convolution (called AddOpt), which improves the utilization of local registers, thereby reducing on-chip cache accesses. Our experiments with a real-world deep CNN, VGG-16, on GPGPU-Sim and Titan X show that the proposed methods, ZeroSkip and AddOpt, achieve 51.8% higher convolution performance than the baseline Winograd convolution. Moreover, even without any hardware modification, AddOpt alone gives 35.6% higher performance on a real hardware platform, Titan X.
Starting Page	1
Ending Page	10
Page Count	10
File Format	PDF
ISBN	9781450344838
DOI	10.1145/2968456.2968476
Language	English
Publisher	Association for Computing Machinery (ACM)
Publisher Date	2016-10-01
Publisher Place	New York
Access Restriction	Subscribed
Subject Keyword	Zero data Data reuse Convolutional neural networks
Content Type	Text
Resource Type	Article

Sl.	Authority	Responsibilities	Communication Details
1	Ministry of Education (GoI), Department of Higher Education	Sanctioning Authority	https://www.education.gov.in/ict-initiatives
2	Indian Institute of Technology Kharagpur	Host Institute of the Project: The host institute of the project is responsible for providing infrastructure support and hosting the project	https://www.iitkgp.ac.in
3	National Digital Library of India Office, Indian Institute of Technology Kharagpur	The administrative and infrastructural headquarters of the project	Dr. B. Sutradhar bsutra@ndl.gov.in
4	Project PI / Joint PI	Principal Investigator and Joint Principal Investigators of the project	Dr. B. Sutradhar bsutra@ndl.gov.in Prof. Saswat Chakrabarti will be added soon
5	Website/Portal (Helpdesk)	Queries regarding NDLI and its services	support@ndl.gov.in
6	Contents and Copyright Issues	Queries related to content curation and copyright issues	content@ndl.gov.in
7	National Digital Library of India Club (NDLI Club)	Queries related to NDLI Club formation, support, user awareness program, seminar/symposium, collaboration, social media, promotion, and outreach	clubsupport@ndl.gov.in
8	Digital Preservation Centre (DPC)	Assistance with digitizing and archiving copyright-free printed books	dpc@ndl.gov.in
9	IDR Setup or Support	Queries related to establishment and support of Institutional Digital Repository (IDR) and IDR workshops	idr@ndl.gov.in

Optimizing memory efficiency for deep convolutional neural networks on GPUs

A Novel Fast Approach for Convolutional Networks with Small Filters Based on GPU

Implementation of energy-efficient fast convolution algorithm for deep convolutional neural networks based on FPGA

Fast image scanning with deep max-pooling convolutional neural networks

Hardware Acceleration of a Generalized Fast 2-D Convolution Method for Deep Neural Networks

CNNdroid: GPU-Accelerated Execution of Trained Deep Convolutional Neural Networks on Android

Deep convolutional neural networks for acoustic modeling in low resource languages

Convolutional Neural Networks for human activity recognition using mobile sensors

Fast Convolution Operations on Many-Core Architectures

Zero and data reuse-aware fast convolution for deep neural networks on GPU

Similar Documents

Optimizing memory efficiency for deep convolutional neural networks on GPUs

A Novel Fast Approach for Convolutional Networks with Small Filters Based on GPU

Implementation of energy-efficient fast convolution algorithm for deep convolutional neural networks based on FPGA

Fast image scanning with deep max-pooling convolutional neural networks

Hardware Acceleration of a Generalized Fast 2-D Convolution Method for Deep Neural Networks

CNNdroid: GPU-Accelerated Execution of Trained Deep Convolutional Neural Networks on Android

Deep convolutional neural networks for acoustic modeling in low resource languages

Convolutional Neural Networks for human activity recognition using mobile sensors

Fast Convolution Operations on Many-Core Architectures

Zero and data reuse-aware fast convolution for deep neural networks on GPU