de
en
Close
Detailsuche
Bibliotheken
Projekt
Imprint
Privacy Policy
Close
Publizieren
Besondere Sammlungen
Digitalisierungsservice
Hilfe
Impressum
Datenschutz
jump to main content
Search Details
Quicksearch:
OK
Result-List
Title
Title
Content
Content
Page
Page
Search Book
MCTS-based approximate accelerator synthesis / by Muhammad Awais. Paderborn, 2021
Content
Acknowledgements
Abstract
Zusammenfassung
Table of Contents
Introduction
Challenges for CMOS technology in the Nano-era
Approximate computing
Approximate computing at different layers of system stack
Approximate arithmetic blocks
The focus of this thesis
Contributions of the thesis
Organization of the remainder
Background and Related Work
Automated synthesis of approximate accelerators
Approximation instance generation
Search space exploration
Quality evaluation
Error metrics
Related works
Search-based methods
Analytical methods
MCTS-based Framework for Approximate Accelerator Synthesis
Chapter overview
Proposed MCTS-based framework for approximate accelerator synthesis
Overview of the framework
MCTS and its adaptation to the accelerator circuit synthesis
Selection policy
Search space expansion policy
Selection of approximate transformations
Reward function
Experimental setup and results
Setup of experiments
Results and discussion
Towards a modular and flexible framework CIRCA
Motivation behind development of CIRCA
Concept and architecture of CIRCA
Input stage
The QUAES stage
The output stage
Search space exploration methods in CIRCA
Hill climbing (HC)
Simulated annealing (SA)
Monte Carlo tree search (MCTS)
Bounding the search space
Experimental setup and results
Setup of experiments
Results
Further Discussion
Chapter conclusion
Hybrid Methodology for Synthesis of Approximate Accelerators
Chapter overview
Motivational example
Proposed hybrid methodology
Analytical bit-width estimation through EVT
Extreme value theory
Analytical bit-width estimation via EVT
Parallel search-based optimization
Parallel MCTS
Experimental results
Setup of experiments
Results and discussion
Chapter conclusion
Machine Learning-based MCTS
Chapter overview
Motivation and background
DNN enabled MCTS-based AxAC synthesis framework
Deep neural network based error estimation regressor models
Design space exploration via MCTS
Experimental results
Setup of experiments
Results and discussion
Chapter conclusion
Conclusion
Summary
Future directions
List of Tables
List of Figures
Author's Publications
Bibliography
The search-operation requires javascript to be activated.