【演講】AIOT智慧物聯網論壇

主辦單位|國立臺灣大學物聯網中心、國立臺灣大學資訊安全技術中心、財團法人資訊工業策進會資安科技研究所、TWISC資通安全研究與教育中心、科技部人工智慧技術暨全幅健康照護聯合研究中心、臺灣大學人工智慧中心
協辦單位|國立臺灣大學電資學院、中華電信、台大國際產學聯盟
時間|018年11年29日(四)9:00-17:00
地點|博理館101演講廳

演講題目與摘要

1.Enhancing External BP Learning with Internal Omnipresent Supervision Training Paradigm: A Systematic Design of Deep Learning Networks
美國普林斯頓大學貢三元教授

Much success of Deep Learning Networks (DLNs) depends on trial-and-error, they all have appeared to be very ad-hoc. It is therefore desirable (if not imperative) to explore a methodical and analytical design and come up with a new learning paradigm. It is well known that there exists huge discrepancy between optimization and generalization, exemplified by four key gaps/issues: data distribution, network capacity, optimization metric, and algorithm. These four issues will serve as the basis our systematic design methodology. In addition, DLNs are vulnerable to various curse-of-depth problems, especially vanishing or exploding gradient. In order to circumvent the depth problem altogether, an Omnipresent Supervision(OS) internal training strategy is proposed. To showcase the advantages of incorporating (internal) OS learning into (external) BP learning, three OS-based learning strategies will be presented: (a) Direct weight updating on each hidden layer; (b) MINDnet: designed to monotonically increase the network's discriminant capability; and (c) {\bf OStrim:} trim the network to become more cost effective in power, storages, and FLOPS.Integration of these new strategies will also be discussed.

2.Coordinated 3D Physical World Exploration from Big Visual Data
美國華盛頓大學黃正能教授

With the huge amount of networked video cameras available everywhere nowadays, such as the statically deployed surveillance cameras or the constantly moving cameras on the vehicles or drones, built upon the success of various deep learning architectures for image-based object detection, segmentation and 2D pose estimation, there is an urgent need of systematic and coordinated mining of the dynamic environment in the 3D physical world, so that the explored information can be exploited for various smart city applications, such as security surveillance, intelligent transportation, business statistics collection, health monitoring of communities, and etc. In this talk, I will first present an automated and robust human/vehicle tracking directly in 3D space through self-calibration of static and moving monocular cameras. The 3D locations and speeds of these tracked objects, as well as their poses, can all be described based on the GPS coordinates, so that the tracked objects from multiple cameras can then be effectively integrated and reconstructed in the 3D real-world space for many smart city and intelligent transportation applications.

3.Life 4.0 through tricorder
國立臺灣大學林清富教授

The advance of science and technology preserves the physical life of human beings, but we also like to have well beings of living life. Hence, we promote life 4.0 that contains health, safety, quality and fashion. To realize life 4.0, new technologies are needed. In particular, our environment is full of pollutants that threaten our life, so a smart and portable device, named Tricoder, becomes necessary. Therefore, handy health-care devices are highly desired by people nowadays in order to well enjoy the benefits of civilization enhanced by science and technology. In this regard, our team combines groundbreaking researches of National Taiwan University that include nano- & micro-technology, photonics technology, IC design, artificial intelligence computing, biomedical sensing, and so on. Our goal is to develop portable I o T+ devices that can detect pollutants and health hazards as well as monitor human health conditions at any place and any time. These health-care portable devices can be widely used in various conditions such as detecting environmental pollution, pesticide residues, fatigue driving causes, real-time temperature, antibody test and physiological status monitor, etc. Above all, this pioneering technology aims to achieve the purpose of health care through environmental and physiological monitoring. This research primarily focuses on two important aspects. One is to detect potential harmful chemicals from the environment. This can be achieved by data transmission, verification, and judgement via I o T networks to provide users timely and correct information as well as appropriate prevention guidelines. The other is that, through these light-weight and portable devices, one can conveniently examine body conditions and obtain important messages to prevent disease or take appropriate actions for health. Furthermore, through the data collection, the could center could accumulate huge amount of information for BIG DATA analysis and give artificial intelligence a broad scope of capability to provide very useful findings to serve human beings.

4.Leveraging Taiwan Open Innovation Infrastructure for AIot
國立臺灣大學王崇智教授

5.AIoT Development and Application
中華電信謝文生經理

活動成果

隨著無線通訊技術的成熟與人工智慧技術的發展,近年人工智慧(AI)生態系與物聯網(I o T)網結合且不斷擴大,逐漸匯流為A I oT,並帶來多項新趨勢。從智慧交通、智慧監控、智慧零售、智慧醫療與智慧工廠等各種 A I o T 裝置軟硬整合解決方案正在各地開枝散葉,在 A I o T 技術驅動下,也帶領創新創意商務新概念。台灣擁有世界一流的高科技人才,搭載世界級頂尖的硬體製造水準,相繼吸引兩大國際企業微軟及亞馬遜雲端運算在台灣設置AI研發中心及聯合創新中心,掌握這一波科技革新將加速帶動台灣產業在物聯網時代的轉型與創新。

林宗男主任主持

圖一:論壇會場

參與觀眾

圖二:與會嘉賓

由左至右為吳沛遠教授、林昌鴻教授、蘇炫榮教授、林宗男教授、貝蘇章老師、貢三元教授、黃正能教授、魏宏宇教授等及其他重要產業界人員共襄盛舉。

參與觀眾

圖三:貢三元教授精彩演講