{"id":{"repo_id":"uiuc","oai_identifier":"oai:www.ideals.illinois.edu:2142/110581"},"canonical_url":"https://search.dev.ndltd.org/etd/uiuc/oai:www.ideals.illinois.edu:2142/110581","repository":{"repo_id":"uiuc","name":"University of Illinois - Urbana-Champaign","base_url":"https://www.ideals.illinois.edu/oai-pmh"},"display":{"title":"TwinDNN: A tale of two deep neural networks","abstract":"Compression technologies for deep neural networks (DNNs), such as weight quantization, have been widely investigated to reduce the model size so that they can be implemented on hardware with strict resource restrictions. However, one major disadvantage of model compression is accuracy degradation. To deal with this problem effectively, we propose a new compressed network inference scheme with a high accuracy but slower DNN coupled with its highly compressed DNN version that typically delivers much faster inference speed but with a lower accuracy. During the inference, we determine the confidence of the prediction of the compressed DNN, and infer the original neural network for the inputs that are considered not confident by the compressed DNN. The proposed design uses a balanced number of resources available on the hardware and can deliver overall accuracy close to the high accuracy model, but with the inference speed closer to the compressed DNN. We demonstrate our design on two image classification tasks: CIFAR-10 and ImageNet. Our experiments show that our design can recover up to 94% of accuracy drop caused by extreme network compression, with more than 90% increase in throughput compared to just using the original DNN. This is more than 17% extra accuracy recovery and 36% extra speedup compared to the previous work with a similar concept on VGG-16. This is the first work that considers using a highly compressed DNN along with the original DNN in parallel to achieve high accuracy and speed at the same time, while maintaining the resource balance by using two different main computation sources on the field programmable gate array (FPGA).","abstract_html":"Compression technologies for deep neural networks (DNNs), such as weight quantization, have been widely investigated to reduce the model size so that they can be implemented on hardware with strict resource restrictions. However, one major disadvantage of model compression is accuracy degradation. To deal with this problem effectively, we propose a new compressed network inference scheme with a high accuracy but slower DNN coupled with its highly compressed DNN version that typically delivers much faster inference speed but with a lower accuracy. During the inference, we determine the confidence of the prediction of the compressed DNN, and infer the original neural network for the inputs that are considered not confident by the compressed DNN. The proposed design uses a balanced number of resources available on the hardware and can deliver overall accuracy close to the high accuracy model, but with the inference speed closer to the compressed DNN. We demonstrate our design on two image classification tasks: CIFAR-10 and ImageNet. Our experiments show that our design can recover up to 94% of accuracy drop caused by extreme network compression, with more than 90% increase in throughput compared to just using the original DNN. This is more than 17% extra accuracy recovery and 36% extra speedup compared to the previous work with a similar concept on VGG-16. This is the first work that considers using a highly compressed DNN along with the original DNN in parallel to achieve high accuracy and speed at the same time, while maintaining the resource balance by using two different main computation sources on the field programmable gate array (FPGA).","abstract_has_math":false,"creators":["Jeong, Hyunmin"],"institution":"University of Illinois at Urbana-Champaign","degree_name":"M.S.","degree_level":"Thesis","degree_discipline":"Electrical & Computer Engr","degree_department":null,"school":null,"contributors":["Chen, Deming"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2021,"date_issued":"2021-09-17T01:13:29Z","date_published":"2021-09-17T01:13:29Z","updated_at":"2026-07-22T22:24:52Z","subjects":["Hardware Accelerator","High-Level-Synthesis","Machine Learning","Neural Network Quantization"],"languages":["en"],"rights":["Copyright 2021 Hyunmin Jeong"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"http://hdl.handle.net/2142/110581","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Chen, Deming"]},{"key":"dc:creator","label":"Author","values":["Jeong, Hyunmin"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2021-09-17T01:13:29Z","2021-04-27","2021-05"]},{"key":"dc:type","label":"Dc Type","values":["text","Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical & Computer Engr"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["M.S."]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Illinois at Urbana-Champaign"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Hardware Accelerator","High-Level-Synthesis","Machine Learning","Neural Network Quantization"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Copyright 2021 Hyunmin Jeong"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["http://hdl.handle.net/2142/110581"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Compression technologies for deep neural networks (DNNs), such as weight quantization, have been widely investigated to reduce the model size so that they can be implemented on hardware with strict resource restrictions. However, one major disadvantage of model compression is accuracy degradation. To deal with this problem effectively, we propose a new compressed network inference scheme with a high accuracy but slower DNN coupled with its highly compressed DNN version that typically delivers much faster inference speed but with a lower accuracy. During the inference, we determine the confidence of the prediction of the compressed DNN, and infer the original neural network for the inputs that are considered not confident by the compressed DNN. The proposed design uses a balanced number of resources available on the hardware and can deliver overall accuracy close to the high accuracy model, but with the inference speed closer to the compressed DNN. We demonstrate our design on two image classification tasks: CIFAR-10 and ImageNet. Our experiments show that our design can recover up to 94% of accuracy drop caused by extreme network compression, with more than 90% increase in throughput compared to just using the original DNN. This is more than 17% extra accuracy recovery and 36% extra speedup compared to the previous work with a similar concept on VGG-16. This is the first work that considers using a highly compressed DNN along with the original DNN in parallel to achieve high accuracy and speed at the same time, while maintaining the resource balance by using two different main computation sources on the field programmable gate array (FPGA).","Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2021-09-16 without embargo terms","The student, Hyunmin Jeong, accepted the attached license on 2021-04-26 at 18:10.","The student, Hyunmin Jeong, submitted this Thesis for approval on 2021-04-26 at 18:17.","This Thesis was approved for publication on 2021-04-27 at 10:49.","DSpace SAF Submission Ingestion Package generated from Vireo submission #16570 on 2021-09-16 at 16:48:26","Made available in DSpace on 2021-09-17T01:13:29Z (GMT). No. of bitstreams: 2 JEONG-THESIS-2021.pdf: 983979 bytes, checksum: 8791b5c96711f6b4e46df61b6a550954 (MD5) LICENSE.txt: 4207 bytes, checksum: dc958dc564ddc0dddbfe3a8d92114645 (MD5) Previous issue date: 2021-04-27"]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["TwinDNN: A tale of two deep neural networks"]}]}],"canonical_facts":{"dc:contributor":["Chen, Deming"],"dc:creator":["Jeong, Hyunmin"],"dc:date":["2021-09-17T01:13:29Z","2021-04-27","2021-05"],"dc:description":["Compression technologies for deep neural networks (DNNs), such as weight quantization, have been widely investigated to reduce the model size so that they can be implemented on hardware with strict resource restrictions. However, one major disadvantage of model compression is accuracy degradation. To deal with this problem effectively, we propose a new compressed network inference scheme with a high accuracy but slower DNN coupled with its highly compressed DNN version that typically delivers much faster inference speed but with a lower accuracy. During the inference, we determine the confidence of the prediction of the compressed DNN, and infer the original neural network for the inputs that are considered not confident by the compressed DNN. The proposed design uses a balanced number of resources available on the hardware and can deliver overall accuracy close to the high accuracy model, but with the inference speed closer to the compressed DNN. We demonstrate our design on two image classification tasks: CIFAR-10 and ImageNet. Our experiments show that our design can recover up to 94% of accuracy drop caused by extreme network compression, with more than 90% increase in throughput compared to just using the original DNN. This is more than 17% extra accuracy recovery and 36% extra speedup compared to the previous work with a similar concept on VGG-16. This is the first work that considers using a highly compressed DNN along with the original DNN in parallel to achieve high accuracy and speed at the same time, while maintaining the resource balance by using two different main computation sources on the field programmable gate array (FPGA).","Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2021-09-16 without embargo terms","The student, Hyunmin Jeong, accepted the attached license on 2021-04-26 at 18:10.","The student, Hyunmin Jeong, submitted this Thesis for approval on 2021-04-26 at 18:17.","This Thesis was approved for publication on 2021-04-27 at 10:49.","DSpace SAF Submission Ingestion Package generated from Vireo submission #16570 on 2021-09-16 at 16:48:26","Made available in DSpace on 2021-09-17T01:13:29Z (GMT). No. of bitstreams: 2 JEONG-THESIS-2021.pdf: 983979 bytes, checksum: 8791b5c96711f6b4e46df61b6a550954 (MD5) LICENSE.txt: 4207 bytes, checksum: dc958dc564ddc0dddbfe3a8d92114645 (MD5) Previous issue date: 2021-04-27"],"dc:format":["application/pdf"],"dc:identifier":["http://hdl.handle.net/2142/110581"],"dc:language":["en"],"dc:rights":["Copyright 2021 Hyunmin Jeong"],"dc:subject":["Hardware Accelerator","High-Level-Synthesis","Machine Learning","Neural Network Quantization"],"dc:title":["TwinDNN: A tale of two deep neural networks"],"dc:type":["text","Thesis"],"thesis:degree_discipline":["Electrical & Computer Engr"],"thesis:degree_level":["Thesis"],"thesis:degree_name":["M.S."],"thesis:institution_name":["University of Illinois at Urbana-Champaign"]},"updated_at":"2026-07-22T22:24:52Z"}