推薦人email address

(沒時間用餐) 衛生福利部雙和醫院(委託臺北醫學大學興建經營) 睡眠中心主任/胸腔內科主治醫師 / 臺北醫學大學 醫學系/呼吸治療學系 副教授 /醫學院睡眠研究中心 研發主任

Mobile Network Resource Management for 5G and Beyond

Centimeter-level Indoor Visible Light Positioning

Abstract: Positioning with visible light in the indoor environments have a number of advantages. Existing lighting infrastructure can be leveraged to save cost, simple optical propagation mechanisms lead to centimeter-level accuracy, and light sensor data can be processed with minimum computational power and energy consumption. In this talk, I will introduce our latest research works on the topic. In the first work, our idea is to retrofit existing luminaries with a special lampshade, generating polarized light to create “colors” that are invisible to human eyes, yet can be perceived by a simple color sensor and leveraged to position the user device very accurately. Another positioning technology, on the other hand, uses addressable LED tubes that replaces existing fluorescent tubes, to provide positioning service. With commodity cameras in many devices, tubes can be identified with simple yet imperceptible spatial patterns on the tube, providing coarse positioning results. Then, more fine-grained position estimation can be calculated with the appearance of the tube in the captured image. I will also show our evaluation results demonstrating centimeter-level accuracy in both works.

Bio:
Hsin-Mu (Michael) Tsai is a Professor in the Computer Science and Information Engineering at National Taiwan University, Taipei, Taiwan. He received his B.S.E in Computer Science and Information Engineering from National Taiwan University and his M.S. and Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University. He spent a year in General Motors Research and Development as an intern researcher during his Ph.D. study. From 2013 to 2015, Dr. Tsai co-led the intelligent transportation system group in the Intel-NTU Connected Context Computing Center, a research center jointly established by Intel, National Taiwan University, and National Science Council, Taiwan, to address research challenges in Internet of Things.

Dr. Tsai's recognitions include 2018 Delta Research Excellence Award, 2017 NTU EECS Academic Contribution Award, 2015 ACM Taipei Chapter’s K. T. Li Young Researcher Award, 2014 Intel Labs Distinguished Collaborative Research Award, 2013 Intel Early Career Faculty Award (the first to receive this honor outside of North America and Europe), and 2013 National Taiwan University's Distinguished Teaching Award (awarded to top 1% teaching faculty at the university). He has served as General co-chair for IEEE Vehicular Networking Conference (VNC) 2018, TPC co-chairs for ACM CarSys 2017, IEEE VNC 2016, ACM VANET 2013, and as a founding co-chair for ACM Visible Light Communication System (VLCS) Workshop in 2014. His research interests include vehicular networking and communications, intelligent transportation systems, and visible light communications and positioning.

Cooperative Distributed Deep Neural Network Deployment over Mobile Networks

bio: Jian-Jhih received the Ph.D. degree in computer science from National Tsing Hua University, Taiwan, in 2014. He is currently an assistant professor in the Department of Computer Science and Information Engineering, National Chung Cheng University, Taiwan. He was a postdoctoral fellow in the Institute of Information Science, Academia Sinica, Taiwan. His research interests include cloud computing, mobile computing, and software-defined networking. He is especially interested in algorithm design for networks.

Synopsys R&D Director, Solutions Group

Deep Learning Acceleration for Embedded Vision

Synopsys新思科技的HR Simin, SiMin.Liao@synopsys.com

賴吉昌先生目前服務於新思科技，負責籌建並領導新竹AI設計中心。在此之前幫安謀國際籌建新竹CPU設計中心，與全球研發團隊合作開發新世代智財產品，並主導創建安謀中國設計團隊。2005年賴先生自美返台參與創建晶心科技公司，並於多年後加入聯發科技技術長辦公室幕僚團隊。期間致力於建立新技術團隊，落實先進軟硬體智財架構設計技術平台，主導前瞻技術策略研究規劃，並在Linaro、OIC、HSA等國際性組織領導跨公司合作。稍早旅居美國時，賴先生專精於計算機結構，資料暨運算安全架構，以及設計驗證等先端核心技術，經歷數家新創團隊建制，曾負責微處理器、MODEM、MCU IP、網路安全等相關產品開發。賴吉昌先生在1990年畢業於國立交通大學控制工程研究所。

以智慧製造為核心應用的高彈性類神經網路架構技術 / High-flexible neural network algorithm and architecture designs for the future Industry 4.0 era

製造業為國家經濟發展的重要指標，隨著人口老化以及人力成本逐年提高的前提下，傳統上透過大量製造來降低生產成本的模式已經逐漸不適用。得利於物聯網、大數據、以及人工智慧技術的蓬勃發展，德國政府於2011年在漢諾威工業展中率先提出工業4.0轉型概念，並開啟了第四次工業革命的浪潮。顯然的，各式工業4.0技術的發展都須仰賴高可靠度的機器運作。然而，由於近年來原物料價格的上漲，也間接造成工廠端針對機器的維運成本逐年提高，造成工廠的獲利不如預期。因此，如何建構一套能夠及時診斷機械健康狀態的預測性維護(predictive maintenance, PdM)系統，以便在機械故障前安排維修，避免意外性的停機已成為現今智慧工廠的熱門議題。在本次演講中，我首先將簡要介紹工業4.0的技術發展趨勢。接著，我將介紹類神經網路技術如何協助在未來工業4.0應用中的預測性維護技術 // Manufacturing is an important indicator of economic development in every country. With the aging of the population and the gradual increase in labor costs, it will become inefficient to reduce the production cost through the model of mass manufacturing. Benefit from the technology booming of IoT, big data, and artificial intelligence, the German government first proposed the concept of the industrial revolution in 2011, which is called Industry 4.0 and it opens the 4th industry revolution on the world. Obviously, each industry 4.0 technology depends on the high-reliable machinery operation significantly. However, the maintenance cost of each factory increases with respect to the increasing cost of raw material, which makes the factory profit worse than expected. Hence, it becomes a hot topic to construct predictive maintenance (PdM) system, which helps to diagnose the health status of the machinery to prevent the unexpected fault. In this talk, I will briefly introduce the development trend of Industry 4.0. Afterward, I will introduce how the neural network technology assists with the PdM system development in the future Industry 4.0 era.

Deep Transfer Learning for Visual Analysis

楊鈞翔 (Lixel CEO), 張凱傑 (Lixel Software RD Manager)

EDRA: Experience Driven Rate Adaptation based on Deep Reinforcement Learning for 802.11ac Networks

The IEEE 802.11ac has become the mainstream Wi-Fi technology that supports gigabit speeds and offers a great number of rate options. Enabling its high-speed features, e.g., 160 MHz bandwidth, can be a challenge to current 802.11ac rate adaptation (RA) solutions, since more rate options impose more overhead on their rate search. It can cause non-scalable RA designs to suffer. In this work, we identify three limitations of current 802.11ac RAs by considering two representative solutions, Iwlwifi and Minstrel, which are Intel and Linux default RAs respectively. We thus propose a scalable and intelligent RA solution, EDRA, which is powered to address these limitations by the deep reinforcement learning (DRL) technique. Its scalability lies in its online learning capability that can automatically derive low-overhead avenues to approach highest-goodput rates by learning from experiences. It does not require any manually observed correlation of important factors and performance, which conventional RAs need. We develop EDRA using the Intel Wi-Fi driver and Google TensorFlow with an asynchronous framework across kernel and user spaces. To the best of our knowledge, we are the first to apply the DRL into a practical RA solution running on commodity 802.11ac devices. Our evaluation results show that EDRA can outperform Iwlwifi and Minstrel by up to 821.4% and 242.8% respectively in various cases.

Chi-Yu Li is currently an Assistant Professor with the Department of Computer Science, National Chiao Tung University (NCTU). He received his PhD degree in computer science from University of California, Los Angeles (UCLA) in 2015. Before joining UCLA, he received his Master and Bachelor degrees from the Department of Computer Science, NCTU. His research interests include wireless networking, mobile networks and systems, and network security. He has published many papers in top-tier academic conferences of both networking and security areas, such as ACM MOBICOM, IEEE INFOCOM, and ACM CCS. He received the Award of MTK Young Chair Professor (2016), MOST Young Scholar Research Award (2017-2020), and the Best Paper Award in IEEE CNS 2018.

maggie@faraday-tech.com

From spec to system: ASIC design service 葫蘆裡賣的是甚麼藥

Hsiang-Yun Cheng(鄭湘筠)

CITI at Academia Sinica

Assistant Research Fellow (助研究員)

Memristor-based In-Memory Computing for Deep Learning

Deep neural networks (DNNs) have grown in prominence in recent years. Their intensive computing and memory demands introduce performance and energy efficiency challenges to the underlying processing hardware. Memristor-based DNN accelerators have been shown to be a promising solution to meet the performance and energy efficiency challenges for DNN inference. In contrast to the conventional von Neumann architecture, where computation and data storage are separated, emerging memristor devices, such as resistive random access memory (ReRAM), are able to perform arithmetic operations beyond data storage. Despite this promising potential, the development of memristor-based DNN accelerators is still in its early stage and there remain challenges to overcome. One primary concern is the errors induced by the non-idealities in current memristor devices. How to efficiently exploit the sparsity in DNN models to save energy is also an important design issue. In this talk, I will introduce our recent studies that aim to overcome these design challenges to enable reliable and energy-efficient memristor-based DNN inferences. I will also share my vision on several future research directions

Hsiang-Yun Cheng is currently an Assistant Research Fellow of the Research Center for Information Technology Innovation (CITI) at Academia Sinica. She received her B.S. and M.S. degree in Computer Science and Information Engineering from National Taiwan University and her Ph.D. degree in Computer Science and Engineering from Pennsylvania State University. Her research falls primarily in the field of computer architecture, with an emphasis on memory system design and domain-specific acceleration. She is especially interested in exploiting emerging technologies and the characteristics of modern applications to design energy-efficient computing systems.

網宇實體系統的設計與應用 / Designs and applications for cyber physical systems

目前軟體定義網路提供了良好的可程式化控制平面. 然而在交換機內部的資料平面還是傳統不可更改的. 因此P4語言被提出來, 可以用來更改交換內部資料 平面處理封包的方式. 包含如何比對封包欄位以及定 義如何處理封包. P4讓網路更有彈性, 有助於未來網 路技術的發展.

lisasung@synology.com

Synology C2 Hybrid Share：巨量雲端系統之開發與挑戰

Time conflict this semester

yes, will confirm date after lunar new year

>> to follow up after Lunar New Year to confirm date

>> did not decline but deferred (indefinitely) due to nCoV virus