Yoshitomo Matsubara
- Machine learning
- Computer vision
- Natural language processing
- Information retrieval
-
2016Master of Applied InformaticsUniversity of Hyogo, Kobe, Japan
-
2014B.E. in Computer and Information ScienceNational Institute of Technology, Akashi College, Japan
He is a Ph.D. Candidate in Computer Science at the University of California, Irvine (UCI), working on computer vision and NLP projects with Profs. Marco Levorato, Sameer Singh, and Stephan Mandt. Before UCI, he has obtained his Master’s and Bachelor’s degrees from the University of Hyogo and the National Institute of Technology, Akashi College, respectively. His Master’s and Bachelor’s theses’ topics were behavioral biometrics, such as keystroke dynamics, and flick authentication. He also has work experience at industry such as Amazon.com Services, Inc., Slice Technologies, Yahoo Japan Corporation, and Recruit Holdings Co., Ltd. for research projects in various domains: NLP, information retrieval, computer vision, and recommender systems.
Publications
No matches
, Head Network Distillation: Splitting Distilled Deep Neural Networks for Resource-Constrained Edge Computing Systems, IEEE Access, vol. 8, pp. 212177 – 212193, 2020. CodeReleaseBibTeX¶ @article{matsubara-callegaro-2020-access, author = {Y. {Matsubara} and D. {Callegaro} and S. {Baidya} and M. {Levorato} and S. {Singh}}, title = {Head Network Distillation: Splitting Distilled Deep Neural Networks for Resource-Constrained Edge Computing Systems}, journal = {IEEE Access}, volume = {8}, number = {}, pages = {212177-212193}, year = {2020}, doi = {10.1109/ACCESS.2020.3039714}, }Edge Computing
Split Computing
Deep Neural Network
Markov Decision Processes
, Optimal Task Allocation for Time-Varying Edge Computing Systems with Split DNNs, in 2020 IEEE Global Communications Conference: Selected Areas in Communications: Internet of Things and Smart Connected Communities (Globecom2020 SAC IoTSCC), Taipei, Taiwan, Dec. 2020. BibTeX¶@inproceedings{callegaro-matsubara-2020-globecom, author = {Davide Callegaro and Yoshitomo Matsubara and Marco Levorato}, title = {Optimal Task Allocation for {Time-Varying} Edge Computing Systems with Split {DNNs}}, booktitle = {2020 IEEE Global Communications Conference: Selected Areas in Communications: Internet of Things and Smart Connected Communities (Globecom2020 SAC IoTSCC)}, address = {Taipei, Taiwan}, month = dec, year = {2020}, days = {6}, keyword = {Edge Computing; Split Computing; Deep Neural Network; Markov Decision Processes}, }, COVIDLies: Detecting COVID-19 Misinformation on Social Media, in EMNLP 2020 NLP COVID-19 Workshop, Oct. 2020. BibTeX¶ @inproceedings{hossain-logan-2020-nlpcovid19, author = {Tamanna Hossain and Robert L. Logan IV and Arjuna Ugarte and Yoshitomo Matsubara and Sean Young and Sameer Singh}, title = {COVIDLies: Detecting COVID-19 Misinformation on Social Media}, booktitle = {EMNLP 2020 NLP COVID-19 Workshop}, month = oct, year = {2020}, }, Citations Beyond Self Citations: Identifying Authors, Affiliations, and Nationalities in Scientific Papers, in Proceedings of the 8th International Workshop on Mining Scientific Publications, Aug. 2020, pp. 9 – 20. BibTeX¶ @inproceedings{matsubara-singh-2020-wosp, author = {Yoshitomo Matsubara and Sameer Singh}, title = {Citations Beyond Self Citations: Identifying Authors, Affiliations, and Nationalities in Scientific Papers}, booktitle = {Proceedings of the 8th International Workshop on Mining Scientific Publications}, pages = {9--20}, month = aug, year = {2020}, }, Reranking for Efficient Transformer-Based Answer Selection, in Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, Jul. 2020, pp. 1577 – 1580. ReleaseAbstractBibTeX¶ IR-based Question Answering (QA) systems typically use a sentence selector to extract the answer from retrieved documents. Recent studies have shown that powerful neural models based on the Transformer can provide an accurate solution to Answer Sentence Selection (AS2). Unfortunately, their computation cost prevents their use in real-world applications. In this paper, we show that standard and efficient neural rerankers can be used to reduce the amount of sentence candidates fed to Transformer models without hurting Accuracy, thus improving efficiency up to four times. This is an important finding as the internal representation of shallower neural models is dramatically different from the one used by a Transformer model, e.g., word vs. contextual embeddings.
@inproceedings{matsubara-vu-2020-sigir, author = {Yoshitomo Matsubara and Thuy Vu and Alessandro Moschitti}, title = {Reranking for Efficient Transformer-Based Answer Selection}, booktitle = {Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval}, pages = {1577--1580}, month = jul, year = {2020}, doi = {10.1145/3397271.3401266}, }, Neural Compression and Filtering for Edge-Assisted Real-Time Object Detection in Challenged Networks, arXiv.org e-Print Archive, Jul. 2020. CodeReleaseAbstractBibTeX¶ The edge computing paradigm places compute-capable devices — edge servers — at the network edge to assist mobile devices in executing data analysis tasks. Intuitively, offloading compute-intense tasks to edge servers can reduce their execution time. However, poor conditions of the wireless channel connecting the mobile devices to the edge servers may degrade the overall capture-to-output delay achieved by edge offloading. Herein, we focus on edge computing supporting remote object detection by means of Deep Neural Networks (DNNs), and develop a framework to reduce the amount of data transmitted over the wireless link. The core idea we propose builds on recent approaches splitting DNNs into sections — namely head and tail models — executed by the mobile device and edge server, respectively. The wireless link, then, is used to transport the output of the last layer of the head model to the edge server, instead of the DNN input. Most prior work focuses on classification tasks and leaves the DNN structure unaltered. Herein, our focus is on DNNs for three different object detection tasks, which present a much more convoluted structure, and modify the architecture of the network to: (i) achieve in-network compression by introducing a bottleneck layer in the early layers on the head model, and (ii) prefilter pictures that do not contain objects of interest using a convolutional neural network. Results show that the proposed technique represents an effective intermediate option between local and edge computing in a parameter region where these extreme point solutions fail to provide satisfactory performance.
@article{matsubara-levorato-2020-icpr, author = {Yoshitomo Matsubara and Marco Levorato}, title = {Neural Compression and Filtering for Edge-Assisted Real-Time Object Detection in Challenged Networks}, journal = {arXiv.org e-Print Archive}, month = jul, year = {2020}, archiveprefix = {arXiv}, eprint = {2007.15818}, primaryclass = {cs.CV}, }, Split Computing for Complex Object Detectors: Challenges and Preliminary Results, in Proceedings of the 4th International Workshop on Embedded and Mobile Deep Learning, Jul. 2020, pp. 7 – 12. AbstractBibTeX¶ Following the trends of mobile and edge computing for DNN models, an intermediate option, split computing, has been attracting attentions from the research community. Previous studies empirically showed that while mobile and edge computing often would be the best options in terms of total inference time, there are some scenarios where split computing methods can achieve shorter inference time. All the proposed split computing approaches, however, focus on image classification tasks, and most are assessed with small datasets that are far from the practical scenarios. In this paper, we discuss the challenges in developing split computing methods for powerful R-CNN object detectors trained on a large dataset, COCO 2017. We extensively analyze the object detectors in terms of layer-wise tensor size and model size, and show that naive split computing methods would not reduce inference time. To the best of our knowledge, this is the first study to inject small bottlenecks to such object detectors and unveil the potential of a split computing approach.
@inproceedings{matsubara-levorato-2020-emdl, author = {Yoshitomo Matsubara and Marco Levorato}, title = {Split Computing for Complex Object Detectors: Challenges and Preliminary Results}, booktitle = {Proceedings of the 4th International Workshop on Embedded and Mobile Deep Learning}, pages = {7--12}, month = jul, year = {2020}, }Information Systems
Mobile Information Processing Systems
Multimedia Streaming
, Distilled Split Deep Neural Networks for Edge-Assisted Real-Time Systems, in Proceedings of the ACM International Conference on Mobile Computing and Networking (MobiCom), Los Cabos, Mexico, Oct. 2019, Workshop on Hot Topics in Video Analytics and Intelligent Edges. CodeReleaseAbstractBibTeX¶Offloading the execution of complex Deep Neural Networks (DNNs) models to compute-capable devices at the network edge, that is, edge servers, can significantly reduce capture-to-output delay. However, the communication link between the mobile devices and edge servers can become the bottleneck when channel conditions are poor. We propose a framework to split DNNs for image processing and minimize capture-to-output delay in a wide range of network conditions and computing parameters. The core idea is to split the DNN models into head and tail models, where the two sections are deployed at the mobile device and edge server, respectively. Different from prior literature presenting DNN splitting frameworks, we distill the architecture of the head DNN to reduce its computational complexity and introduce a bottleneck, thus minimizing processing load at the mobile device as well as the amount of wirelessly transferred data. Our results show 98% reduction in used bandwidth and 85% in computation load compared to straightforward splitting.
@inproceedings{matsubara-baidya-2019-mobicom, author = {Yoshitomo Matsubara and Sabur Baidya and Davide Callegaro and Marco Levorato and Sameer Singh}, title = {Distilled Split Deep Neural Networks for Edge-Assisted Real-Time Systems}, booktitle = {Proceedings of the ACM International Conference on Mobile Computing and Networking (MobiCom)}, note = {Workshop on Hot Topics in Video Analytics and Intelligent Edges}, address = {Los Cabos, Mexico}, month = oct, year = {2019}, doi = {10.1145/3349614.3356022}, keyword = {Information Systems, Mobile Information Processing Systems, Multimedia Streaming}, }, Screen Unlocking by Spontaneous Flick Reactions With One-Class Classification Approaches, in Proceedings of the IEEE International Conference on Machine Learning and Applications (ICMLA), Anaheim, California, Dec. 2016, pp. 752 – 757. ReleaseBibTeX¶ @inproceedings{matsubara-nishimura-2016-icmla, author = {Yoshitomo Matsubara and Haruhiko Nishimura and Toshiharu Samura and Hiroyuki Yoshimoto and Ryohei Tanimoto}, title = {Screen Unlocking by Spontaneous Flick Reactions With One-Class Classification Approaches}, booktitle = {Proceedings of the IEEE International Conference on Machine Learning and Applications (ICMLA)}, pages = {752--757}, address = {Anaheim, California}, month = dec, year = {2016}, doi = {10.1109/icmla.2016.0134}, }, Robustness of Keystroke Dynamics Against Changing in Input and Recognition Conditions for Japanese Atypical Text, in Proceedings of the International Symposium on Technology for Sustainability (ISTS), Bangkok, Thailand, Nov. 2012, pp. 103 – 106. BibTeX¶ @inproceedings{matsubara-samura-2012-ists, author = {Yoshitomo Matsubara and Toshiharu Samura and Haruhiko Nishimura}, title = {Robustness of Keystroke Dynamics Against Changing in Input and Recognition Conditions for Japanese Atypical Text}, booktitle = {Proceedings of the International Symposium on Technology for Sustainability (ISTS)}, pages = {103--106}, address = {Bangkok, Thailand}, month = nov, year = {2012}, }