Advanced search options

Advanced Search Options 🞨

Browse by author name (“Author name starts with…”).

Find ETDs with:

in
/  
in
/  
in
/  
in

Written in Published in Earliest date Latest date

Sorted by

Results per page:

You searched for id:"oai:etd.ohiolink.edu:ucin1552391639148868". One record found.

Search Limiters

Last 2 Years | English Only

No search limiters apply to these results.

▼ Search Limiters

1. George, Abhinav Kurian. Fault tolerance and re-training analysis on neural networks.

Degree: MS, Engineering and Applied Science: Computer Engineering, 2019, University of Cincinnati

In the current age of big data, artificial intelligence and machine learning technologies havegained much popularity. Due to the increasing demand for such applications, neural networksare being targeted toward hardware solutions. Owing to the shrinking feature size, number ofphysical defects are on the rise. These growing number of defects are preventing designers fromrealizing the full potential of the on-chip design. The challenge now is not only to find solutionsthat balance high-performance and energy-efficiency but also, to achieve fault-tolerance of acomputational model. Neural computing, due to its inherent fault tolerant capabilities, canprovide promising solutions to this issue. The primary focus of this thesis is to gain deeperunderstanding of fault tolerance in neural network hardware.As a part of this work, we present a comprehensive analysis of fault tolerance by exploringeffects of faults on popular neural models: multi-layer perceptron model and convolution neuralnetwork. We built the models based on conventional 64-bit floating point representation. Inaddition to this, we also explore the recent 8-bit integer quantized representation. A faultinjector model is designed to inject stuck-at faults at random locations in the network. Thenetworks are trained with the basic backpropagation algorithm and tested against the standardMNIST benchmark. For training pure quantized networks, we propose a novel backpropagationstrategy. Depending on the performance degradation, the faulty networks are re-trained torecover their accuracy.Results suggest that: (1) neural networks cannot be considered as completely fault tolerant;(2) quantized neural networks are more susceptible to faults; (3) using a novel training algorithmfor quantized networks, comparable accuracy is achieved; (4) re-training is an effectivestrategy to improve fault tolerance. In this work, 30% improvement in quantized network isachieved as compared to 6% improvement in floating point networks using the basic backpropagationalgorithm. We believe that using more advanced re-training strategies can enhance faulttolerance to a greater extent. Advisors/Committee Members: Jone, Wen-Ben (Committee Chair).

Subjects/Keywords: Computer Engineering; neural networks; artificial intelligence; fault tolerance; quantization

Record DetailsSimilar RecordsGoogle PlusoneFacebookTwitterCiteULikeMendeleyreddit

APA · Chicago · MLA · Vancouver · CSE | Export to Zotero / EndNote / Reference Manager

APA (6th Edition):

George, A. K. (2019). Fault tolerance and re-training analysis on neural networks. (Masters Thesis). University of Cincinnati. Retrieved from http://rave.ohiolink.edu/etdc/view?acc_num=ucin1552391639148868

Chicago Manual of Style (16th Edition):

George, Abhinav Kurian. “Fault tolerance and re-training analysis on neural networks.” 2019. Masters Thesis, University of Cincinnati. Accessed July 20, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1552391639148868.

MLA Handbook (7th Edition):

George, Abhinav Kurian. “Fault tolerance and re-training analysis on neural networks.” 2019. Web. 20 Jul 2019.

Vancouver:

George AK. Fault tolerance and re-training analysis on neural networks. [Internet] [Masters thesis]. University of Cincinnati; 2019. [cited 2019 Jul 20]. Available from: http://rave.ohiolink.edu/etdc/view?acc_num=ucin1552391639148868.

Council of Science Editors:

George AK. Fault tolerance and re-training analysis on neural networks. [Masters Thesis]. University of Cincinnati; 2019. Available from: http://rave.ohiolink.edu/etdc/view?acc_num=ucin1552391639148868

.