課程大綱

課程資訊

課程名稱	專題討論 Seminar
開課學期	112-1
授課對象	電機工程學研究所
授課教師	劉智弘
課號	EE8001
課程識別碼	921 D0040
班次	03
學分	0.0
全/半年	半年
必/選修	必修
上課時間	星期一7(14:20~15:10)
上課地點	明達231
備註	計算機科學組。博士班必修。限博士班總人數上限：150人

課程簡介影片
核心能力關聯	核心能力與課程規劃關聯圖
課程大綱
為確保您我的權利,請尊重智慧財產權及不得非法影印
課程概述
課程目標
課程要求
預期每週課後學習時數
Office Hours
指定閱讀
參考書目
評量方式 (僅供參考)

課程進度

週次	日期	單元主題
第1週	9/4	Regulation
第2週	9/11	Jason Chou (CTO, CMoney Technology Corporation) Give the sense of landing AI applications - Especially ChatGPT? Large Language Model(LLM)? generative AI? - How to take the advantages of this AI waves - Share my experience in deploying AI applications - What I have learnt from my failed cases
第3週	9/18	Meng-Tsung Tsau (Assistant Research Fellow, Institute of Information Science, Academia Sinica) Title: Graph Computations Using Sublinear Space Abstract: When tackling graph problems involving real-world data, we often encounter a common challenge: input sizes exceed our machine's memory capacity. In such scenarios, textbook algorithms might not have efficient implementations, necessitating alternative algorithm designs. During this presentation, we will explore the concept of sparse certificates, enabling us to perform some classical graph computations as if the entire dataset resides in memory. In addition, we will learn how to identify a problem that does not admit a sparse certificate. Finally, a number of research problems suited for exploration by graduate students will be introduced.
第4週	9/25	吳柏鋒 (台灣大學物理系) 主題：Data, Open Data, and Big Data in Modern Astronomy 摘要：自21世紀初始，現代天文學開始迎來大轉變。數項大型巡天計畫搜集了海量資料並將其公開，使得資料探勘成為一大趨勢。天文學社群享受到公開資料的益處，持續將更多資源挹注於巡天觀測計畫以及建設開放資料的架構，讓資料更多，更好，更容易使用。現今的天文學者不必然需要懂得操作望遠鏡，也可對尖端研究做出貢獻。海量的開放資料也自然地讓機器學習及深度學習成為常見的資料分析方法。我會介紹過去20年天文學在海量資料觀測、資料探勘、資料分析上的發展，以及2030年代，下一個世代的大型計畫們面臨的挑戰。
第5週	10/02	麥德倫(臺灣大學生命科學系) 主題：Computational Insights into Biology and Medicine: from Proteins to DNA 摘要： In this talk, I will share the confluence of computational science and the life sciences, elucidating our research contributions across three topics. Firstly, we conducted a proteome-wide network analysis based on our proposed clustering method, and provided an evolutionary relationship among the proteins, with a specific emphasis on their implications for cancer progression and fundamental physiological functions. Second, we developed a multi-step methodology coupled with the construction of a mathematical model grounded in physical principles to unravel the intricate folding processes of membrane proteins, which constitute a substantial subset of potential drug targets. Lastly, we conducted a comprehensive multi-omics data analysis aimed at exploring the regulatory aspects of circRNAs and their interactions with genomic variants, microRNAs, and risk genes in autism. In summary, my presentation will cover a spectrum from proteins to DNA, showcasing the transformative potential of computational methodologies in advancing our understanding of biological and medical phenomena.
第7週	10/16	Hee-Kap Ahn (Pohang University of Science and Technology) Title: Optimal Paths in the Presence of Obstacles Abstract: In the shortest path problem, we seek to find a path of minimum length between two points under various settings. The shortest path between two nodes in a graph is a path connecting them such that the sum of the weights of its constituent edges is minimized. The shortest path between two points in a simple polygon is a polygonal chain of minimum length that is contained in the polygon, and it is unique. There can be more than one shortest path in the prese
第9週	10/30	De-Nian Yang Title: Diffusion Optimization for Multimedia Social Networks Abstract: Diffusion takes on different forms across various network applications. During an epidemic crisis, viruses propagate through the physical contact of individuals within a contact network. Analyzing this network can help control the epidemic's spread. In e-commerce, information is disseminated through social interactions within social networks, enabling the potential for viral marketing by harnessing social influence. In the context of traffic engineering for VR videos, the transmission of network resources occurs over communication networks, making it imperative to optimize transmission costs to enhance user experience. In this talk, 1) for epidemic control, we aim to optimize the spread of the epidemic while maintaining user satisfaction by scheduling containment actions for different groups of people and by timely recommending Points of Interest (POIs) to users, ensuring that the spread of the epidemic remains controlled. 2) In e-commerce, we factor in users' dynamic personal perceptions and consider the ripple effect on user preferences, social influence, and item associations to promote multiple items across various promotions. 3) For VR video traffic engineering, we focus on online multicast combined with view synthesis by utilizing multi-tree and multi-slot rerouting to minimize the overall transmission cost. For the above problems, we design new algorithms infused with innovative ideas, including 1) analyzing infection ratios, contact risks and recovered ratios for containment actions and POIs selection and scheduling, 2) identifying various target markets and evaluating dynamic reachability of items to determine promoting orders, and 3) evaluating view popularity-cost ratio for view and source selection while jointly selecting synthesis nodes and routing for dynamic join and leave. We conduct experiments on real datasets (e.g., Haslemere, Manhattan, Douban, Gowalla, Cogent, and Xspedius) to evaluate the performance of the proposed algorithms.
第10週	11/6	Shih-Han Hung Title: Applications of Quantum Supremacy Experiments Abstract: After three decades of research in quantum computing theory and experiments, the world has seen the emergence of noisy quantum devices capable of solving special sampling problems using 50-60 qubits or approximately 100 photons in a way that is believed to achieve quantum computational supremacy, i.e., the quantum devices outperform any existing classical computer for these tasks. Despite these achievements, whether they lead to practical applications has remained unclear. In this talk, I will discuss the latest developments in quantum supremacy proposals. I will introduce my research on new cryptographic applications that build upon these proposals, including certified randomness from untrusted quantum devices, classical verification of devices with limited quantum depth, and verification of full-power quantum computers. These novel applications challenge us to further explore the limitations of quantum devices, and repurpose various supremacy proposals for practical uses.
第11週	11/13	何宗諭（博士生）題目 : Edge Computing with Machine Learning 摘要 : 我目前在國內的網通大廠擔任研發，本部門性質是將一些先進技術導入公司現有的產品上，讓公司的產品產生加值加價。面對的難題往往不是技術本身，而是如何在有限的硬體資源下，將先進的人工智慧演算法導入限制的設備中，本次的分享將介紹過去實作的案例，如DGA detection, traffic classification, 和 QoS 這幾個經驗分享，介紹如何使用AI提升產品資安防禦能力，進而優化客戶的使用者體驗。陳羿翔（博士生）題目：Robust Vehicle Similarity Learning in Real Hazy Scenes Based on Semi-supervised Learning. 惡劣氣候下的車輛搜索方法摘要： Re-identification (ReID) 是一項智慧都市相關應用，雖然已有多數方法在公開資料集獲得良好數據，但他們忽略了真實世界當中惡劣天氣導致的影像破壞，使得研究方法與真實世界應用存在嚴重的Domain Gap，除此之外，惡劣氣候下的數據往往難以收集與註記，因此我們開發新穎的演算法，結合Image Restoration與Semi-supervised相關知識進行訓練，並成功取得最佳成績，提升現有辨識模型於真實世界的泛用性。
第15週	12/11	王瀚緯（博士生）題目：利用iPhone Pro/iPad Pro的光學雷達與相機來達成各種醫學測量的Apps 摘要：利用iPhone Pro/iPad Pro的光學雷達與相機並安裝自製的各種應用程式，來做到各種醫學測量應用。 1. 測量身高：一群人可以同時測量身高並即時顯示在頭頂，特點是部分身體被其他人擋住也能量。 2. 褥瘡面積：自動辨識褥瘡範圍並算出面積。 3. 燒燙傷面積:自動辨識燒燙傷範圍並算出面積與燒燙傷比例（利用辨識出來的手掌面積換算）。 4. 跌倒偵測：隨時隨地用手機偵測畫面內是否有人跌倒，並即時透過email通知，特點是可以分辨跌倒在床上/沙發上（不發警報）或在地上（發警報）。陳璽文（博士生）題目：How can neural network understand graph algorithm? 摘要：Recently, neural network has gained a great success in understating the high dimensional data, e.g., image or audio. However, there are limited studies about expressiveness of neural network to enable a more reliable machine learning system. In this talk, I would like to share our recent effort in how neural network can learn the graph algorithm, and thus providing a theoretical guarantee for neural network serving as a general agent in the real-word applications.