*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Energy-efficient Digital Hardware Platform for Learning Complex Systems
Committee:
Dr. Saibal Mukhopadhyay, ECE, Chair , Advisor
Dr. Sudhakar Yalamanchili, ECE
Dr. Arijit Raychowdhury, ECE
Dr. Sek Chai, SRI International
Dr. Hyesoon Kim, CS
Abstract:
System learning is the most fundamental research area in engineering domain. It is a modeling method to map external inputs to the corresponding outputs with/without physically analyzing the system between them. The system can be simple enough, e.g. a linear time-invariant system, to be easily identified by a simple mathematical model. However, it can be a more complex system, such as a nonlinear dynamic system, which is highly difficult to understand with mathematical representations.
In this presentation, energy-efficient digital hardware to understand the wide range of complex systems using different approaches will be presented. As a data-driven approach, several neural network algorithms are selected for the system learning. The focused system is related to vision tasks such as image or video processing. Several design algorithms and analysis to realize low-power neural network accelerators will be discussed. The proposed low-power design methods are not limited to certain tasks, but are based on algorithmic analysis for general applicability. For a model-based approach, the cohesive method of frequency-domain analysis and interpolation will be presented to simplify learning process on the thermal behavior of integrated circuits. Also, a programmable hardware for simulating dynamical systems will be presented. The proposed platform accelerates the computation of solving a wide class of differential equations.
