2020 year in review by McKinsey

McKinsey, 2020 year in review, 2020/12/28.

動態規劃 (dynamic programming)

考慮多階段決策的問題，例如下方的最短路徑問題，需要從起點 A、經過中間站的 BCDE、走到終點 F；A 到 B 的距離為 1，D 到 F 的距離為 100 等等。

現在的問題很小，可以使用窮舉法 (method of enumeration) 找出所有的可能。走 ABDF 總距離 102、ABEF 總距離 8、ACDF 總距離 106、ACEF 總距離 9，所以最短路徑為 ABEF。

如何解決一般化的問題？首先考慮短視的策略  (myopic policy)，站在 A 點時，在 B 和  C 兩條路中選擇 B (AB = 1 < 4 = AC)；站在 B 點時，選擇 D；站在 D 點時，只能選擇 F。總成本為 102，路徑為 ABDF。

You Can’t Escape Hyperparameters and Latent Variables

Charles Isbell, You Can’t Escape Hyperparameters and Latent Variables: Machine Learning as a Software Engineering Enterprise, keynote speech at NeuIPS, Dec 8th, 2020. (slide 70: What have we learned?)

Successful technological fields have a moment when they become pervasive, important, and noticed. They are deployed into the world and, inevitably, something goes wrong. A badly designed interface leads to an aircraft disaster. A buggy controller delivers a lethal dose of radiation to a cancer patient. The field must then choose to mature and take responsibility for avoiding the harms associated with what it is producing. Machine learning has reached this moment. In this talk, I will argue that the community needs to adopt systematic approaches for creating robust artifacts that contribute to larger systems that impact the real human world. I will share perspectives from multiple researchers in machine learning, theory, computer perception, and education; discuss with them approaches that might help us to develop more robust machine-learning systems; and explore scientifically interesting problems that result from moving beyond narrow machine-learning algorithms to complete machine-learning systems.

這樣的傅斯年

沈珮君，這樣的傅斯年（上），聯合報，2020-08-06

最近，重回傅鐘下，發現一塊小碑，是陳維昭校長在民國91年立的，寫了傅鐘21響意義，「一天只有二十一小時，剩下三小時是用來沉思的」。...

他在民國38年1月20日接掌台大，39年12月20日去世，還差一個月才滿兩年。他太累了，台大千頭萬緒，上自典章制度、攬才招生，下至親筆給學生、工人回信，細大不捐，曾有一個外文系校友去信給他，抱怨畢業一年卻仍無畢業證書，他親自洽問，才知教部因要查核每人是否修滿學分，一人不符，全體不發，光復以後即未發過畢業證書，在他要求下才解決。他去世前幾天，跟朱家驊說：「你把我害了，台大的事真是多，吃不消，我的命要斷送在台大了。」他死在省參議會，報告完台大校務，緩緩回到座位，臉發白，手奇冷，血壓一度高到230，五小時後即逝。

《中央日報》刊出他在省參議會的最後發言，傅斯年強調台大獎學金制度不應廢止，「對於那些資質好、肯用功的，僅只為了沒錢而不能升學的青年，我是萬分同情的，我不能讓他們被摒棄於校門之外」，這成了他的遺言。當初受惠的台大青年，應都九十幾歲了。...

Mathematics for Machine Learning

Marc Peter Deisenroth, A. Aldo Faisal, and Cheng Soon Ong, Mathematics for Machine Learning, Cambridge University Press, 2020. (pdf)

The book assumes the reader to have mathematical knowledge commonly covered in high school mathematics and physics. For example, the reader should have seen derivatives and integrals before, and geometric vectors in two or three dimensions. Starting from there, we generalize these concepts. Therefore, the target audience of the book includes undergraduate university students, evening learners and learners participating in online machine learning courses.

台灣軟體業的發展

 科技島讀，Ep.126 難產的台灣獨角獸｜特別來賓 XREX 創辦人黃耀文，2020/12/20

台灣獨角獸遲遲難產，讓人擔憂輝煌的半導體與電子製造業後繼無人。相較於以色列、芬蘭、烏克蘭、新加坡等在軟體領域各據地盤，台灣軟體業至今無法破繭而出，更啟人疑竇。島讀邀請成功創業家黃耀文，一起從產品定位、資本、人才等面向，分析台灣新創的錯過與值得。

慈善捐款

 Nicholas Kulish, Giving Billions Fast, MacKenzie Scott Upends Philanthropy, NYT, Dec. 20, 2020.

By disbursing her money quickly and without much hoopla, Ms. Scott has pushed the focus away from the giver and onto the nonprofits she is trying to help. They are the types of organizations — historically Black colleges and universities, community colleges and groups that hand out food and pay off medical debts — that often fly beneath the radar of major foundations.

“If you look at the motivations for the way women engage in philanthropy versus the ways that men engage in philanthropy, there’s much more ego involved in the man, it’s much more transactional, it’s much more status driven,” said Debra Mesch, a professor at the Women’s Philanthropy Institute at Indiana University. “Women don’t like to splash their names on buildings, in general.”

How To Speak by Patrick Winston

Patrick Winston, How To Speak, MIT OpenCourseWare, 2019/12/20.

Feedback Control Perspectives on Learning

Jeff Shamma, Feedback Control Perspectives on Learning, keynote speech at NeuIPS, Dec 8th, 2020.

The impact of feedback control is extensive. It is deployed in a wide array of engineering domains, including aerospace, robotics, automotive, communications, manufacturing, and energy applications, with super-human performance having been achieved for decades. Many settings in learning involve feedback interconnections, e.g., reinforcement learning has an agent in feedback with its environment, and multi-agent learning has agents in feedback with each other. By explicitly recognizing the presence of a feedback interconnection, one can exploit feedback control perspectives for the analysis and synthesis of such systems, as well as investigate trade-offs in fundamental limitations of achievable performance inherent in all feedback control systems. This talk highlights selected feedback control concepts—in particular robustness, passivity, tracking, and stabilization—as they relate to specific questions in evolutionary game theory, no-regret learning, and multi-agent learning.

Listening to Prof. Shamma's talk is always enjoyable and a great learning experience.  

The information on slide 3 by 2 prominent researchers in control:

K. Astrom, "Automatic Control – a Perspective," UON FEBE, 2019/9/5.

G. Stein, “Respect the unstable,” IEEE Control System Magazine, vol. 23, no. 4, pp. 12–25, Aug. 2003. (Cartoons for slides 97 - 98)

 

An AI development platform for industrial systems

Kyle Wiggers, Microsoft launches Project Bonsai, an AI development platform for industrial systems, Venture Beat, May 19, 2020.

Microsoft announced the public preview of Project Bonsai, a platform for building autonomous industrial control systems, during its Build 2020 online conference. The company also debuted an experimental platform called Project Moab that’s designed to familiarize engineers and developers with Bonsai’s functionality.

Project Bonsai is a “machine teaching” service that combines machine learning, calibration, and optimization to bring autonomy to the control systems at the heart of robotic arms, bulldozer blades, forklifts, underground drills, rescue vehicles, wind and solar farms, and more. Control systems form a core component of machinery across sectors like manufacturing, chemical processing, construction, energy, and mining, helping manage everything from electrical substations and HVAC installations to fleets of factory floor robots. But developing AI and machine learning algorithms atop them — algorithms that could tackle processes previously too challenging to automate — requires expertise....

耶魯最受歡迎的金融通識課

 陳志武，耶魯最受歡迎的金融通識課，今周刊，2019

全書文字通俗易懂，沒有公式和金融模型，卻能從歷史角度和量化分析視角來幫助讀者建立金融思維，以經濟眼光看待這個世界的運轉。是一本幫助普通大眾認清財富本源，學會用金融思維看懂世界的金融通識書。

25 頁

金融的核心任務，是要解決人與人之間的跨期價值交換的問題。…… 這些跨期價值交換涉及人與人之間的跨期承諾 (intertemporal commitment)，而跨期承諾是人類社會最難解決的挑戰。

在金融市場出現之前，人類多次的文化和社會組織創新，目的都是為了解決跨期承諾的挑戰，進而提升人與人之間跨期交換的安全度。

在書裡你會學到，傳統習俗、迷信宗教愛情婚姻家庭禮尚往來，以及儒家、基督教等文化的背後，其實含有豐富的金融邏輯。也就是說，許多文化的內涵實際上是因為金融市場的缺點而產生的，是為了解決本來應該由金融出面的問題而來的。

Beat the Slots in Pokémon Using Reinforcement Learning

 Daniel Saunders, How I Beat the Slots in Pokémon Using Reinforcement Learning, towardsdatascience, 2020/12/10. (Python code)

Given a set of possible actions (“arms” of a multi-armed bandit — in this case different machines to try), Thompson sampling optimally trades off exploration vs exploitation to find the best action, by trying the promising actions more often, and so getting a more detailed estimate of their reward probabilities. At the same time, it’s still randomly suggesting the others from time to time, in case it turns out one of them is the best after all. At each step, the knowledge of the system, in the form of posterior probability distributions, is updated using Bayesian logic. The simplest version of the one-armed bandit problem involves Bernoulli trials, where there are only two possible outcomes, reward or no reward, and we are trying to determine which action has the highest probability of reward.

This is a nice article and it should be interesting to teach this example to motivate the students. 

聰明學統計的 13 又 1/2 堂課

愛荷譯，聰明學統計的 13 又 1/2 堂課 : 每個數據背後都有戲, 搞懂才能做出正確判斷，先覺，2013

C.J. Wheelan, Naked Statistics:  Stripping the Dread from the Data, W.W. Norton, 2013. 

 《聰明學經濟的12堂課》作者查爾斯．惠倫再次以他豐富的學養、幽默的風格，以及化繁為簡的無上功力，拋開統計學枯燥生硬的理論，讓你從生活中的大小事，輕鬆搞懂最關鍵實用的統計概念，做出正確決策。他說，本書是要讓最重要的統計觀念變得更合乎直覺，也更容易上手。他說，用統計說謊很容易，沒有統計要找出真相卻很困難。你將從這本書中，得到超乎你想像的重要訊息！而且，是用一種充滿趣味的方式！

適合當課程的補充教材。 

為什麼是 13 又 1/2 堂呢？大概和美國的大學學制有關，學期制一學期 15 週，扣掉考試差不多。

246 頁，作者舉例說明高爾夫課課程對平均桿數的影響，線性迴歸分析當然沒辦法描述。我們需要的是非線性迴歸 (第七章)。

秋野芒劇團

 連翊君，為偏鄉孩子而演！許子漢的「許伯大夢」，大人社團，2018-07-19 

秋野芒劇團創辦人、花蓮東華大學華文系（原中文系）副教授許子漢今年54歲，朋友暱稱他「大伯」，他有個「許伯大夢」，要做6個故事相連的兒童劇，一年演出120場，包括花蓮等地偏鄉，讓國小學童每年看一部戲，直到畢業。...

Artificial Intelligence — The Revolution Hasn’t Happened Yet (人工智慧 -- 革命尚未發生)

Michael Jordan, Artificial Intelligence — The Revolution Hasn’t Happened Yet, Harvard Data Science Review, Apr 19, 2018.

When my spouse was pregnant 14 years ago, we had an ultrasound. There was a geneticist in the room, and she pointed out some white spots around the heart of the fetus. “Those are markers for Down syndrome,” she noted, “and your risk has now gone up to one in 20.” She let us know that we could learn whether the fetus in fact had the genetic modification underlying Down syndrome via an amniocentesis, but amniocentesis was risky—the chance of killing the fetus during the procedure was roughly one in 300. Being a statistician, I was determined to find out where these numbers were coming from. In my research, I discovered that a statistical analysis had been done a decade previously in the UK in which these white spots, which reflect calcium buildup, were indeed established as a predictor of Down syndrome. I also noticed that the imaging machine used in our test had a few hundred more pixels per square inch than the machine used in the UK study. I returned to tell the geneticist that I believed that the white spots were likely false positives, literal white noise.

She said, “Ah, that explains why we started seeing an uptick in Down syndrome diagnoses a few years ago. That’s when the new machine arrived.”

We didn’t do the amniocentesis, and my wife delivered a healthy girl a few months later, but the episode troubled me, particularly after a back-of-the-envelope calculation convinced me that many thousands of people had gotten that diagnosis that same day worldwide, that many of them had opted for amniocentesis, and that a number of babies had died needlessly. The problem that this episode revealed wasn’t about my individual medical care; it was about a medical system that measured variables and outcomes in various places and times, conducted statistical analyses, and made use of the results in other situations. The problem had to do not just with data analysis per se, but with what database researchers call provenance—broadly, where did data arise, what inferences were drawn from the data, and how relevant are those inferences to the present situation? While a trained human might be able to work all of this out on a case-by-case basis, the issue was that of designing a planetary-scale medical system that could do this without the need for such detailed human oversight....

SIAM Conference on Mathematics of Data Science (MDS20)

 SIAM Conference on Mathematics of Data Science (MDS20)  Topological Data Analysis of Complex High-Dim. Layout Configurations for IC Physical Designs

計算機程式教學

需要助教幫忙學生上機實習的原因與方法 

一下時，我的計程補考 (blog)。當時老師的教學方式是板書上課後給作業；對初學者而言，困難的是程式的邏輯 (if/for) 和除錯 (debug)，邏輯可以借由上課中的例子說明之 (blog)，但是除錯肯定是需要幫忙的。國內外許多課程都有類似的設計，例如元智資管的程設、交大應數的高等微積分，連頂尖的 MIT 都有 (課程, 說明)，更何況是一般的學生。

Does Advertising Actually Work?

 Freakonomics, 441. Does Advertising Actually Work? (Part 2: Digital)   (Part 1: TV)

行銷的關鍵是要確定因果和相關，例如某大零售商一年只有在三個重要節日做廣告，銷售量的增加是因為季節效應、還是廣告？在這個有趣的節目裡面，和 eBay 合作的經濟學家，做了一個 A/B 測試實驗，說明其搜尋引擎關鍵字廣告效果有限。當時的 eBay 總經理因此決定縮減一億美元的線上廣告預算。

T. Blake, C. Nosko, and S. Tadelis, Consumer heterogeneity and paid search effectiveness: A large‐scale field experiment, Econometrica, 83 (1), pp. 155-174, 2015.

Our experiments show that the effectiveness of SEM is small for a well-known company like eBay and that the channel has been ineffective on average....

We then use our experimental methods that control for endogeneity to find a ROI of -63%, with a 95% confidence interval of [-124%; -3%], rejecting the hypothesis that the channel yields positive returns at all.

專題實作和金門行銷

這學期上一門大三的專案實作，今天談到以考試為主的傳統教學缺點，我舉兩個例子說明。朋友的小孩去年唸小二，學校教月的陰晴圓缺，竟然是用背誦的，小朋友痛苦不堪。朋友是工程碩士，透過兩個球體和觀察，了解月形狀的變化，輕輕鬆鬆就記下來了。令人驚訝的不是背不下來的小孩，而是班上九成的同學硬背下來，這才是教育的根本問題。更深層的原因是我們的師資培育方式，我們現在是包班制，這是制度造成的問題；數理比較抽象，在芬蘭需要具備有(理工) 碩士才能任教。

矽谷創業之神陳五福的創投心法練成術

連以婷 ，「坦白說，我開始走創投是失敗的！」矽谷創業之神陳五福的創投心法練成術，科技新報，2020 年 02 月 28 日

坦白說我開始走創投是失敗的，因為做投資時還是依循過去當創業家的心態，但這兩者其實很不一樣。創業家要非常樂觀，因為每個人都知道創業成功的機率可能十之一二，如果不保持樂觀的心態、不相信現在做的事情會成功，別說做下去連創業都不敢想。

但做創投的人則要倒過來，要非常的謹慎、要抽絲剝繭找出所有可能的失敗風險，與創業家互相制衡並補足他們不懂的地方，提醒他們需要小心的部分。我一開始做創投就過於同理這些創業家，好像把過去那份創業的激情投射在他們身上，於是就跟著他們樂觀，反而缺少了制衡力。...

Uber 應對失誤的 A/B 測試

 地球圖輯隊，道歉也有公式可學？他教會Uber如何應對失誤，簡單一招比千百句對不起更有效，2020.11.16   

當天晚上，身為Uber首席經濟學家的里斯特立刻打電話給當時的Uber執行長卡拉尼克（Travis Kalanick），劈頭直說：「這趟旅程爛透了，我再也不會用你的APP了。而且最糟糕的是，我連一句道歉都沒收到。」

The BAIR Blog

The Berkeley Artificial Intelligence Research (BAIR) blog

The BAIR Blog provides an accessible, general-audience medium for BAIR researchers to communicate research findings, perspectives on the field, and various updates. Posts are written by students, post-docs, and faculty in BAIR, and are intended to provide relevant and timely discussion of research findings and results, both to experts and the general audience. Posts on a variety of topics studied at BAIR will appear approximately once every two weeks.

They could explain the technical details in a nice and amazingly clear way, so I really enjoy their writing. Taking this blog as an example, they use two-way consistency and its corresponding picture to explain sparse graphical memory for robust planning. 

The Stoic Challenge (堅忍的挑戰)

William B. Irvine, The Stoic Challenge: A Philosopher's Guide to Becoming Tougher, Calmer, and More Resilient, W. W. Norton & Company, 2019.

A practical, refreshingly optimistic guide that uses centuries-old wisdom to help us better cope with the stresses of modern living.

Researchers develop machine-learning optimizer to slash product design costs

Brett Hansard, Researchers develop machine-learning optimizer to slash product design costs, Argonne National Laboratory,  NOVEMBER 16, 2020.

Speed up the product design optimization process: 

It employs a novel machine learning technique that helps users focus on how to most efficiently target computational resources. (Machine learning is an application of artificial intelligence that allows systems to automatically learn and improve from experience.)

"ActivO runs the simulations in a very smart way and quickly identifies the parts of the design space we should focus on," explained Pal. "A process that used to take two to three months to give you the optimum design can now be completed within about a week."

馬友友演奏念故鄉

馬友友是我非常喜愛的一位音樂家，這一次來台灣開音樂會，二話不說，馬上買票進場，也算圓了夢想。德弗乍克 (Antonín Leopold Dvořák) 的念故鄉是他的安可曲，觸動心弦；上網找了一下，有之前的錄影 (音樂從 2:26 開始)。

馬大康：未來AI技術將有7大挑戰與新機會

余至浩，新任Google臺灣董事總經理馬大康：現階段全球AI發展仍處於石器時代，未來AI技術將有7大挑戰與新機會，iThome，2020-11-15

但講到現今AI的發展，馬大康仍以AI石器時代來加以形容，現階段全球AI發展仍在起步階段，未來還有很大成長空間。他並歸納出7項未來AI技術挑戰與新機會。

躁動的帝國

 林添貴譯，躁動的帝國：從清帝國的普世主義，到中國的民族主義，一部250年的中國對外關係史，八旗文化，2020

Odd Arne Westad, Restless Empire: China and the World since 1750,  Basic Books, 2015.

談中國近代史，因為國共兩黨、不同族群，產生了不一樣的史觀。作者是挪威的歷史學家，曾在北京大學和清華大學教學，從國際局勢和各方角力，提供了不同的視野，值得深讀，以增加我們看事情的廣度。24 頁的註釋資料，引用許多重要政治人物所說的話，和官方歷史教科書的說法有出入。

在這一場巨大的轉變中，除了檯面上的政治人物外，書中也道盡一般老百姓的無奈、痛苦、和生離死別。

學習數學的四個層次：(4) 純粹滿足好奇心或求知慾

學習數學的四個層次：(0) 如何學數學，(1) 代表具備基礎的知識與能力，(2) 邏輯推理和抽象思考的能力，(3) 在許多行業的應用，(4) 純粹滿足好奇心或求知慾。

2015/12/1 初稿，持續更新中。

一般性說明

• 因為內在的動力與好奇心，許多數學家 (和許多研究人員一樣) 終其一身，都在解決一些未知的問題。
• 許多看似沒用的數學理論，後來卻發現有許多有趣且重要的應用。

Deep Learning: An Introduction for Applied Mathematicians

Catherine F. Higham and Desmond J. Higham, Deep Learning: An Introduction for Applied Mathematicians, SIAM Review, 2019, Vol. 61, No. 4, pp. 860–891.

A Nonparametric Approach to Modeling Choice with Limited Data

Vivek F. Farias, Srikanth Jagabathula, and Devavrat Shah, A Nonparametric Approach to Modeling Choice with Limited Data, Management Science, February 2013, Vol. 59, No. 2, pp. 305-322. 

Choice models today are ubiquitous across a range of applications in operations and marketing. Real-world implementations of many of these models face the formidable stumbling block of simply identifying the “right” model of choice to use. Because models of choice are inherently high-dimensional objects, the typical approach to dealing with this problem is positing, a priori, a parametric model that one believes adequately captures choice behavior. This approach can be substantially suboptimal in scenarios where one cares about using the choice model learned to make fine-grained predictions; one must contend with the risks of mis-specification and overfitting/underfitting. Thus motivated, we visit the following problem: For a “generic” model of consumer choice (namely, distributions over preference lists) and a limited amount of data on how consumers actually make decisions (such as marginal information about these distributions), how may one predict revenues from offering a particular assortment of choices? An outcome of our investigation is a nonparametric approach in which the data automatically select the right choice model for revenue predictions. The approach is practical. Using a data set consisting of automobile sales transaction data from a major U.S. automaker, our method demonstrates a 20% improvement in prediction accuracy over state-of-the-art benchmark models; this improvement can translate into a 10% increase in revenues from optimizing the offer set. We also address a number of theoretical issues, among them a qualitative examination of the choice models implicitly learned by the approach. We believe that this paper takes a step toward “automating” the crucial task of choice model selection.

The authors formulated the minimum revenue problem under consumer choices as a linear programming with exponential growing of decision variables in terms of product number. Based on duality, they developed polynomial-time algorithms by using constraint sampling and efficient representation of purchase permutations. Profs. Farias and Shah then founded the company Celect and was later acquired by Nike. Once again, it demonstrates the positive cycle of advanced research and academic-industrial collaboration.

Superhuman AI for Multiplayer Poker

 Noam Brown, EC'20: Superhuman AI for Multiplayer Poker

Toward a Blend of Machine Learning and Microeconomics

Michael I. Jordan on the Future of AI and Machine Learning (Full Video) | ODSC West 2019

 

創業之國以色列

徐立妍譯，創業之國以色列：教育思維X兵役制度X移民政策X創投計畫，打造建國七十年成長50倍的經濟奇蹟，木馬文化，2017

Dan Senor and Saul Singer, Start-up Nation: the story of Israel’s economic miracle, Twelve, 2011.

以色列是全世界創投最興盛的國家之一，即使鄰國猛烈轟炸、網路經濟泡沫破滅等危機，以色列的創投資金依然不受影響，且持續成長。新加坡和杜拜亦試圖複製以色列模式，為什麼不成功？到底，以色列有什麼祕訣，在這麼多不利的條件下，仍然表現亮眼？

50 years of Data Science

David Donoho, 50 Years of Data Science, Journal of Computational and Graphical Statistics, Volume 26, Issue 4, 2017, Pages 745-766.

More than 50 years ago, John Tukey called for a reformation of academic statistics. In “The Future of Data Analysis,” he pointed to the existence of an as-yet unrecognized science, whose subject of interest was learning from data, or “data analysis.” Ten to 20 years ago, John Chambers, Jeff Wu, Bill Cleveland, and Leo Breiman independently once again urged academic statistics to expand its boundaries beyond the classical domain of theoretical statistics; Chambers called for more emphasis on data preparation and presentation rather than statistical modeling; and Breiman called for emphasis on prediction rather than inference. Cleveland and Wu even suggested the catchy name “data science” for this envisioned field. A recent and growing phenomenon has been the emergence of “data science” programs at major universities, including UC Berkeley, NYU, MIT, and most prominently, the University of Michigan, which in September 2015 announced a $100M “Data Science Initiative” that aims to hire 35 new faculty. Teaching in these new programs has significant overlap in curricular subject matter with traditional statistics courses; yet many academic statisticians perceive the new programs as “cultural appropriation.” This article reviews some ingredients of the current “data science moment,” including recent commentary about data science in the popular media, and about how/whether data science is really different from statistics. The now-contemplated field of data science amounts to a superset of the fields of statistics and machine learning, which adds some technology for “scaling up” to “big data.” This chosen superset is motivated by commercial rather than intellectual developments. Choosing in this way is likely to miss out on the really important intellectual event of the next 50 years. Because all of science itself will soon become data that can be mined, the imminent revolution in data science is not about mere “scaling up,” but instead the emergence of scientific studies of data analysis science-wide. In the future, we will be able to predict how a proposal to change data analysis workflows would impact the validity of data analysis across all of science, even predicting the impacts field-by-field. Drawing on work by Tukey, Cleveland, Chambers, and Breiman, I present a vision of data science based on the activities of people who are “learning from data,” and I describe an academic field dedicated to improving that activity in an evidence-based manner. This new field is a better academic enlargement of statistics and machine learning than today’s data science initiatives, while being able to accommodate the same short-term goals.

川湖回收水、防空污連閃兩危機

侯良儒，中小企業也能部署！川湖回收水、防空污連閃兩危機，商業周刊第1709期，2020-08-13

不接受砍價決心轉型

從代工轉做品牌，先投入無毒電鍍

 

處理廢水選最難的路 

花二千萬建回收廠， 遇旱保住訂單

當時 ，川湖只要停工一天 ，就是一千二百萬元的損失 ；同時，假如無法出貨 ，客戶便會轉向中國的工廠下單。

成本和風險的抉擇，還發生在五年前

 

警覺空汙法規將趨嚴 

精算長期成本後，換成天然氣鍋爐

Behaviour Suite for Reinforcement Learning by DeepMind

Ian Osband, Yotam Doron, Matteo Hessel, John Aslanides, Eren Sezener, Andre Saraiva, Katrina McKinney, Tor Lattimore, Csaba Szepesvari, Satinder Singh, Benjamin Van Roy, Richard Sutton, David Silver, Hado Van Hasselt, Behaviour Suite for Reinforcement Learning,  ICLR 2020. (code)

This paper introduces the Behaviour Suite for Reinforcement Learning, or bsuite for short. bsuite is a collection of carefully-designed experiments that investigate core capabilities of reinforcement learning (RL) agents with two objectives. First, to collect clear, informative and scalable problems that capture key issues in the design of general and efficient learning algorithms. Second, to study agent behaviour through their performance on these shared benchmarks. To complement this effort, we open source this http URL, which automates evaluation and analysis of any agent on bsuite. This library facilitates reproducible and accessible research on the core issues in RL, and ultimately the design of superior learning algorithms. Our code is Python, and easy to use within existing projects. We include examples with OpenAI Baselines, Dopamine as well as new reference implementations. Going forward, we hope to incorporate more excellent experiments from the research community, and commit to a periodic review of bsuite from a committee of prominent researchers.

和和機械 5 年前關鍵布局

林洧楨 ，亞洲切彎管機王揭5年前關鍵布局：我一猶豫就死了！ 41歲黑手廠挖台積電人才 和和怎麼躲過工具機衰退？，商業周刊，2020-10-15

這家成軍41年的老工具機廠，專攻金屬管材的切割、打洞、折彎的加工設備，在業界最出名的就是，從美國波音飛機、俄羅斯米格戰鬥機航太管材，到義大利法拉利跑車的排氣管加工都要找它。根據同業推估，它過去5年營收從12億元成長到去年近18億元，專攻高利潤的客製化專用機，讓它營業毛利率超過3成以上。

但其實，5年前它還面臨汽車業訂單驟減的衝擊。...

Important Papers That Were Rejected (Several Times) ...

FIONA MACDONALD, 8 Scientific Papers That Were Rejected Before Going on to Win a Nobel Prize, Science Alert, 19 AUGUST 2016.

Peer-review involves having a group of independent researchers read every paper submitted to a journal to make sure that the methods and conclusions are solid. They will often suggest revisions to be made, and can reject a paper if they think more work needs to be done, or if it's not the right fit for the journal.

Following rejection, the end product is usually better than it would have been originally - or it at least, ends up in a more appropriate journal.

Joshua S. Gans and George B. Shepherd, How Are the Mighty Fallen: Rejected Classic Articles by Leading Economists, The Journal of Economic Perspectives, Vol. 8, No. 1 (Winter, 1994), pp. 165-179.

This paper presents a selection of dispatches from the publication battlefront. We begin by discussing rejections that winners of the Nobel Prize and John Bates Clark Medal have endured, and some other notable cases. We then turn to the record of John Maynard Keynes' quirky refusals, when he was the Economics Journal's editor, of several important articles and authors. Finally, we offer some thoughts about the implications of these findings.

Gloria by A. Vivaldi

 A. Vivaldi - Gloria - Mustonen - Barrocade - Voces Musicales

當亞馬遜遇見Frontier，時尚新裝從300天縮短到7天上架

陳怡如  ，全球最大數位紡織雲來了！當亞馬遜遇見Frontier，時尚新裝從300天縮短到7天上架！，遠見，2020-10-05

以前一件衣服從設計到打樣，需要花45天，300天後上架，最後有40％到80％的衣服賣不出去；轉為數位設計後，只要1天打樣、7天上架，甚至達到零庫存、零浪費的完美生產境界。...

事實上布片數位化並非新鮮事，只是拍照建檔時得考慮大小、顏色、光影、紋路、拍照設備等一連串問題，目前業界提供布片數位化的公司，光是每塊布片要調校機器拍攝，一塊布就得花上50分鐘，一張還要收費30美元。

為了解決棘手的數位化問題，趙均埔花了六年時間，開發四代系統才成功。Frontier的方式很簡單，紡織廠不用添購昂貴設備，只要用公司裡最常見的事務機掃描布片，三分鐘就能上傳完成。接著再透過雲端上多達12個AI引擎，自動辨識布種、克重範圍、紋路、規格、色號，省去手動建檔的功夫。

原先系統辨識一塊布料的時間需要15分鐘，現在透過AWS的雲端資源助攻，辨識時間大幅縮短為40秒。2019年趙均埔登上有著「服飾業最創新論壇」之稱的PI Apparel舞台，底下坐著UA、Nike、Target等一線品牌大咖，從此讓Frontier一炮而紅。

The Deep Learning - Applied Math Connection

 SIAM Mathematics of Data Science (MDS20) Distinguished Lecture Series: Yann LeCun (slide)

A few useful things to know about machine learning

Pedro Domingos, A Few Useful Things to Know About Machine Learning, Communications of the ACM, 2012, Vol. 55, No. 10, Pages 78-87.

This article summarizes 12 key lessons that machine learning researchers and practitioners have learned. These include pitfalls to avoid, important issues to focus on, and answers to common questions. 

Table 1 The three components of learning algorithms.

Learning = Representation + Evaluation + Optimization 

This is a nice overview article which could be assigned as an entry-level course reading. For a teacher, you could cover these topics or provide enough background material in your course so that the students could explore the content by themselves. 

Statistical Modeling: The Two Cultures

Leo Breiman, Statistical Modeling: The Two Cultures, Statistical Science, 2001, Vol. 16, No. 3, 199–231.

There are two cultures in the use of statistical modeling to reach conclusions from data. One assumes that the data are generated by a given stochastic data model. The other uses algorithmic models and treats the data mechanism as unknown. The statistical community has been committed to the almost exclusive use of data models. This commitment has led to irrelevant theory, questionable conclusions, and has kept statisticians from working on a large range of interesting current problems. Algorithmic modeling, both in theory and practice, has developed rapidly in fields outside statistics. It can be used both on large complex data sets and as a more accurate and informative alternative to data modeling on smaller data sets. If our goal as a field is to use data to solve problems, then we need to move away from exclusive dependence on data models and adopt a more diverse set of tools.

Introducing data science

Davy Cielen, Arno D. B. Meysman, and Mohamed Ali, Introducing data science: Big data, machine learning, and more, using Python tools, Manning, May 2016.

Introducing Data Science explains vital data science concepts and teaches you how to accomplish the fundamental tasks that occupy data scientists. You'll explore data visualization, graph databases, the use of NoSQL, and the data science process. You'll use the Python language and common Python libraries as you experience firsthand the challenges of dealing with data at scale. Discover how Python allows you to gain insights from data sets so big that they need to be stored on multiple machines, or from data moving so quickly that no single machine can handle it. This book gives you hands-on experience with the most popular Python data science libraries, Scikit-learn and StatsModels. After reading this book, you'll have the solid foundation you need to start a career in data science.

Data Science in Production: Building Scalable Model Pipelines with Python

Ben Weber, Data Science in Production: Building Scalable Model Pipelines with Python, Independently published, 2020.

Putting predictive models into production is one of the most direct ways that data scientists can add value to an organization. By learning how to build and deploy scalable model pipelines, data scientists can own more of the model production process and more rapidly deliver data products.

Decisive actions to emerge stronger in the next normal

Kevin Sneader, Shubham Singhal, and Bob Sternfels, What now? Decisive actions to emerge stronger in the next normal, McKinsey & Company, September 2020 (pdf)

1. Think of the return as a muscle
2. Focus on high-impact actions
3. Rebuild for speed
4. Reimagine the workforce from the top down
5. Make bold portfolio moves
6. Reset technology plans
7. Rethink the global footprint
8. Take the lead on climate and sustainability
9. Think about the role of regulation and government
10. Make purpose part of everything

台灣引興怎麼堅持豐田管理的零庫存

曾如瑩、管婺媛，工具機天王遇斷鏈潮，怎麼堅持豐田管理的零庫存，商業周刊，2020 年 08 月 31 日

邱奕嘉問（以下簡稱邱）：每次有重大天災，精實管理就會被拿出來檢討。因為低庫存者斷鏈，造成損失，反倒有庫存者，業績成長。經過這次疫情，你認為零庫存概念是否應該調整？

王慶華答（以下簡稱王）：豐田汽車當然（曾經）因為天災而斷鏈，但它也是全世界恢復最快的。企業講究長期利益，不是講短期利益的。斷貨（鏈）是事實沒錯，有存貨者，可能在疫情爆發這 3 個月中活得比別人好，但就是贏這 3 個月，之後呢？即便賺到比別人多一倍的利潤，也只有短期，庫存總會用完。

Linear Algebra and Optimization for Machine Learning: A Textbook

Charu C. Aggarwal, Linear Algebra and Optimization for Machine Learning: A Textbook, Springer, 1st ed, 2020.

This textbook introduces linear algebra and optimization in the context of machine learning. Examples and exercises are provided throughout the book. A solution manual for the exercises at the end of each chapter is available to teaching instructors. This textbook targets graduate level students and professors in computer science, mathematics and data science. Advanced undergraduate students can also use this textbook. The chapters for this textbook are organized as follows: 

Taming the Tail: Adventures in Improving AI Economics

Martin Casado and Matt Bornstein, Taming the Tail: Adventures in Improving AI Economics, a16z.com, August 12, 2020.

The New Business of AI (and How It’s Different From Traditional Software)

 Martin Casado and Matt Bornstein, The New Business of AI (and How It’s Different From Traditional Software), a16z.com, February 16, 2020.

We are huge believers in the power of AI to transform business: We’ve put our money behind that thesis, and we will continue to invest heavily in both applied AI companies and AI infrastructure. However, we have noticed in many cases that AI companies simply don’t have the same economic construction as software businesses. At times, they can even look more like traditional services companies. In particular, many AI companies have:

1. Lower gross margins due to heavy cloud infrastructure usage and ongoing human support;
2. Scaling challenges due to the thorny problem of edge cases;
3. Weaker defensive moats due to the commoditization of AI models and challenges with data network effects....

 Building, scaling, and defending great AI companies – practical advice for founders 

• Eliminate model complexity as much as possible. 
• Choose problem domains carefully – and often narrowly – to reduce data complexity. 
• Plan for high variable costs.
• Embrace services.
• Plan for change in the tech stack. 
• Build defensibility the old-fashioned way. 

Nice blog by Yassine Ouali

 Yassine Ouali

Deep Semi-Supervised Learning

CVPR 2020: A Snapshot 

ECCV 2020: Some Highlights

台灣本土火箭 要讓太空旅行夢想成真

 吳宗信，台灣本土火箭 要讓太空旅行夢想成真，TEDxTaipei，2015 (吳教授的台語真棒)

英業達使用 5G 讓 AI 取代人眼審查

王郁倫，光學檢測聰明10倍！直擊英業達伺服器基地，5G如何讓AI取代人眼審查？，數位時代，2020.08.28

2020年8月中，英業達架設起5G企業專網，利用上傳100Mbps及下載1Gbps的高速，串連生產車間上的AOI（自動光學檢測，Automated Optical Inspection）系統，不僅人力安排減少9成，產線直通率（FPY, First Pass Yield）更拉高至85％。...

差別在於，AOI 告別「單站」智慧，改採「集中」智慧。

AI在學習分辨產品瑕疵時，必須累積足夠多的掃描圖檔，才能精準判讀。5G導入前，每條產線AOI各別利用旁邊的電腦運算判斷，搜集的圖庫也只來自該產線；導入5G後，10條產線AOI，都直接把圖檔上傳雲端集中判斷，數據充足精準度自然提升。

英業達企業電腦事業群全球營運中心副總閻承隆解釋，因為Wi-Fi不穩定，過去智慧工廠每條產線自建AI學習，AOI掃描產品後的圖檔送到產線伺服器運算判別良劣，圖片不夠多精準度就低，要經過不斷學習，精準度才拉高到98％。

英國製造：國家如何維繫經濟命脈

 蔡明燁譯，英國製造：國家如何維繫經濟命脈，立緒，2017

當經濟不斷受挫之際，市場上很少出現正面的經濟觀。產業外移、房市泡沫、物價齊漲，薪資水平卻長期低迷甚至倒退，除了籠統歸因大環境不景氣之外，人們也開始質疑自己究竟能夠產製或銷售什麼有價值的東西，而未來的經濟又該走向何方？

他山之石，可以攻錯，在思考我們國家的產業發展時，或許可以看看英國如何度過金融海嘯，重新調整經濟體質，站穩腳步迎向國際新局的例子。而隨著篇章開展，讀者亦能逐漸將書中分析套用在台灣的經濟發展上，當在面對國內經濟轉型的各種挑戰時，不再如無頭蒼蠅般惶惶不安。

作者伊凡．戴維斯為英國經濟學者，也是長期深入觀察當地產業的財經記者，書中對英國經濟的分析採取正向樂觀的論點，但絕不盲目，而是就事論事，從嚴謹的數據與比較分析中得出持平而論的根據，並做出有力的提醒和檢討，目的是要說服讀者，一個正常開放國家的謀生實力，其實比我們所想像中要強得多。

Matrix Methods in Data Analysis, Signal Processing, and Machine Learning

Gilbert Strang. 18.065 Matrix Methods in Data Analysis, Signal Processing, and Machine Learning. Spring 2018. Massachusetts Institute of Technology: MIT OpenCourseWare, https://ocw.mit.edu. License: Creative Commons BY-NC-SA. (book)

Strang 教授教這門課的時候 83 歲，真的是終身學習的好典範。在美國，這種對專業的執著 (Tapley 教授) 令人欽佩。另外一個例子是 Breiman 教授，71 歲投稿隨機森林 (Random forests)，成為經典論文；其他幾篇重要論文，大都在 65 歲以後以單一作者發表！ 

Programming for the Puzzled

Srini Devadas, Programming for the Puzzled: Learn to Program While Solving Puzzles, The MIT Press, 2017.  

Learning programming with one of “the coolest applications around”: algorithmic puzzles ranging from scheduling selfie time to verifying the six degrees of separation hypothesis.

Rob Smedley From Formula 1 Talks About Using AWS to Improve the Fan Experience

AWS re:Invent 2019 – Rob Smedley From Formula 1 Talks About Using AWS to Improve the Fan Experience, 2019/12/4

Formula 1 has been using Amazon EC2 for Computational Fluid Dynamics (CFD) to simulate race car aerodynamics, achieving the performance of a super computer at a much lower cost and reducing simulation time by an average of 70% — from 60 hours down to 18 hours. With the CFD project, Formula 1 used over 500 million data points to study downforce loss when two vehicles race in close proximity. (A car’s downforce increases its tire grip and cornering speed and reduces lap time.) Based on its CFD simulations, Formula 1 has designed a car for the 2021 racing season that reduces downforce loss in wheel-to-wheel racing from 50% to 15% — and offers a more exciting experience for fans.

More information.

A prosthetic leg that learns from the user's motion could help amputees walk more naturally

Andrew Ng, A prosthetic leg that learns from the user's motion could help amputees walk more naturally, The Batch, August 05, 2020.

What’s new: Researchers from the University of Utah designed a robotic leg that uses machine learning to generate a human-like stride. It also helps wearers step over obstacles in a natural way.

Learning Spark: Lightning-Fast Data Analytics

 Jules S. Damji, Brooke Wenig, Tathagata Das, and Denny Lee, Learning Spark: Lightning-Fast Data Analytics, 2nd edition, 2020, O’Reilly Media. (code)

We welcome you to the second edition of Learning Spark. It’s been five years since the first edition was published in 2015, originally authored by Holden Karau, Andy Konwinski, Patrick Wendell, and Matei Zaharia. This new edition has been updated to reflect Apache Spark’s evolution through Spark 2.x and Spark 3.0, including its expanded ecosystem of built-in and external data sources, machine learning, and streaming technologies with which Spark is tightly integrated.

Over the years since its first 1.x release, Spark has become the de facto big data unified processing engine. Along the way, it has extended its scope to include support for various analytic workloads. Our intent is to capture and curate this evolution for readers, showing not only how you can use Spark but how it fits into the new era of big data and machine learning. Hence, we have designed each chapter to build progressively on the foundations laid by the previous chapters, ensuring that the content is suited for our intended audience....

Most of the examples in the chapters are written in Scala, Python, and SQL. Where necessary, we have infused a bit of Java. 

 The ebook is available for download once you fill in your information at Databrick. 

Jim Keller 為英特爾開出的藥方

工程師在波特蘭，Jim Keller來了，2020 年 08 月 12 日

JK的第一個改革非常符合邏輯, 簡單來說就是兩個重點: IP re-use (重複使用), 還有在IP部門的開發時程和產品部門的整合時程上盡可能的重疊. 他下達的新指令就是, IP team以後不負責hardening, 由產品部門負責, 但是IP team要確保IP是可以很容易的驗證 (verifiable), 而且介面要很乾淨. 這樣一來產品部門可以在很早期就開始驗證, 然後因為hardening統一由產品部門負責, 所以操作條件也一致, 實作起來也比較有效率. 為了完成這個任務, JK在他自己加入五個月後, 從外面挖來了有個人師徒情誼的Netspeed的CEO Sundari Mitra來負責統整所有IP方面的業務.

數位轉型全攻略

黃俊堯，數位轉型全攻略：虛實整合的 WHAT，WHY 與 HOW，商業周刊，2019

不論領域、大公司、中小企業都在談數位轉型，但想要轉跟怎麼轉永遠是兩回事，不要讓數位轉型成為你們公司的痛點。

台大商學院教授黃俊堯現身說法，分享數位轉型的眉角！

The Next Normal: How companies and leaders can reset for growth beyond coronavirus

McKinsey, The Next Normal: How companies and leaders can reset for growth beyond coronavirus, 2020.

A Study of More Than 250 Platforms Reveals Why Most Fail

David B. Yoffie, Annabelle Gawer, and Michael A. Cusumano, A Study of More Than 250 Platforms Reveals Why Most Fail, HBR, May 29, 2019.

Platforms have become one of the most important business models of the 21st century. In our newly-published book, we divide all platforms into two types:  Innovation platforms enable third-party firms to add complementary products and services to a core product or technology. Prominent examples include Google Android and Apple iPhone operating systems as well as Amazon Web Services. The other type, transaction platforms, enable the exchange of information, goods, or services. Examples include Amazon Marketplace, Airbnb, or Uber.

台灣連鎖企業坪效王

未來流通研究所，年度30強，誰是台灣連鎖企業坪效王，2020 / 08 / 04

「坪效」是實體通路有別於其他行業別最特殊的經營指標，也是攸關連鎖企業經營績效的關鍵數據。在全球實體消費市場遭逢疫情衝擊以及電子商務爆發性成長的當下，觀測各類型實體通路企業的坪效指標，不僅可做為企業經營健全強度的分析與衡量基準，也能夠為疫情後台灣實體通路的變革思維帶來一些線索。

未來流通研究所團隊抓取2019年台灣連鎖實體通路企業的全年營收、店鋪數量、營運坪數等數據，彙整成為連鎖企業坪效指標，並羅列前30強企業名單供讀者參考。從名單中可以看出，餐飲業為進榜家數最多的產業別，且當中涵蓋餐館、中式速食、日式速食、西式速食、咖啡飲料店等各種類型，而台灣餐飲業知名品牌鼎泰豐的坪效表現更是大幅領先，穩坐台灣連鎖企業坪效王的制霸地位。

崴昊科技的工程最佳化軟體

李家同，為台灣加油打氣專欄 (213) 國產工程最佳化軟體，2020 年 7 月 29 日

這次要介紹的是一種台灣自行開發的工程最佳化軟體，這種軟體是 CAE (Computer Aided Engineering) 軟體的一種。要設計一個工業產品，通常都要有一個模擬軟體(simulation software)，也就是說，我們要測試一下所設計的產品能否使用。比方說，我們設計了一個馬達，當然要測試這個馬達能不能轉，這可以用模擬軟體來測驗。如果我們設計了一個電子電路，要知道這個電子電路是否符合要求，也可以用模擬軟體來測驗。 

Princeton Consultants Inc.

看到上一篇的新聞，得知他們和 Princeton Consultants 合作。基本上，該公司是使用最佳化以管理和解決企業流程的複雜問題，此方法在歐美企業有廣泛的應用，而且有很好的成效，可以參考他們的顧問項目和合作的公司。 

這一家顧問公司的經營團隊經驗豐富，President and CEO Steve Sashihara 寫了一本書，也有 11 位博士，包含 Dr. Irv Lustig；代表此公司是知識密集的產業，也有能力吸引和保有一流的人才。

聘請的員工來自各種學校和專業，可見他們的教育訓練非常的好。也有其他的優點，例如廣泛的背景和領域知識，有助於和顧客溝通，知道導入企業所需的產業知識，降低溝通成本；另外，多元的學術背景也可以提供解決問題時所需的不同角度和思維，有助於團隊合作。

這樣的管理和用人方式，值得我們思考和學習。

How Birchbox Transformed its Operations With Mathematical Optimization

Gurobi, Webinar – How Birchbox Transformed its Operations With Mathematical Optimization,

July 28th - 29th, 2020

Birchbox – the trailblazing subscription box service that launched in 2010 – had created a mixed-integer programming formulation to determine the products that would be sent to its subscribers in individual boxes on a monthly basis. The goal is to produce a set of different box configurations that are then assigned to customers – so that Birchbox can meet the diverse needs of its varied customer base. 

As the business grew, the mixed-integer program was taking days to solve, and it was impossible to experiment with different business requirements to determine the best set of configurations. The RIP (Reciprocating Integer Programming) technique was created to reduce solution times to typically under 20 minutes using Gurobi, which has dramatically changed the way that Birchbox can run its subscription business.

貝佐斯寫給股東的信 (The Bezos Letters)

李芳齡，貝佐斯寫給股東的信：亞馬遜14條成長法則帶你事業、人生一起飛，大塊文化 ，2019
Steve Anderson and Karen Anderson, The Bezos Letters: 14 Principles to Grow Your Business Like Amazon, Morgan James Publishing, 2019

濃縮21封貝佐斯致股東信精華，構建四階段成長循環，總結14條事業成長法則，揭開貝佐斯透過哪些教訓、心態和步驟，塑造出亞馬遜今日的偉大成就。 

不論組織型態、規模大小、行業別等，任何企業主、領導人、執行長、經理人和員工個人，都可以應用這些法則，快速地讓自己的事業變得更有效率、更有生產力、更成功。

ML and RL for Combinatorial Optimization

Yoshua Bengio, Andrea Lodi, Antoine Prouvost, Machine Learning for Combinatorial Optimization: a Methodological Tour d'Horizon, arXiv:1811.06128v2.

Nina Mazyavkina, Sergey Sviridov, Sergei Ivanov, Evgeny Burnaev, Reinforcement Learning for Combinatorial Optimization: A Survey, arXiv:2003.03600v1.

新創用 ai 找出外觀看不到的標籤，幫電商做到精準「神推薦」

陳君毅，要消費者塞爆購物車！新創用ai找出外觀看不到的標籤，幫電商做到精準「神推薦」，數位時代，2020.07.06

Rosetta.ai Blog

如何打造一個「頂級數據科學團隊」？

TechOrange，如何打造一個「頂級數據科學團隊」？ Linkedin：狂用 A/B Test 就對了！，2020/07/09 (本文經合作夥伴 大數據文摘 授權轉載，並同意 TechOrange 編寫導讀與修訂標題，原文標題為 〈數據科學家”或許不再性感，但“數據團隊”的產業化才剛開始 〉。)

那麼到底如何讓數據的價值最大化呢？從團隊運作方式、商業影響力設定和社會責任等角度，許亞給出了 LinkedIn 的答案：「嵌入式工作，中心化管理」，數據科學團隊更加「專業化」、「工程化」。 

和多數網路服務公司一樣，LinkedIn 的數據科學團隊規模也在近幾年飛速增長。許亞表示，僅是近兩年來，LinkedIn 的數據團隊擴張了近一倍，從 150 人增加到目前的 300 多人。 

許亞提到的數據團隊是指 LinkedIn 中心化的數據科學部門。 如果用一句話來概括 LinkedIn 的中心數據科學團隊的運作方式，那就是「嵌入式工作，中心化管理」。 

和國內不少互聯網公司將數據分析師歸屬於業務 BU、向業務主管匯報不同，LinkedIn 的數據科學團隊成員由許亞的中心部門統籌。雖然在項目工作上，數據科學家們依然會在工位分佈和職能上與業務部門緊密聯繫，但是從職級從屬上，都直接向許亞匯報，不同領域的數據科學家在工作中會有交集，還會一起開會。...

ESG now a third of MioTech’s A.I. business

MioTech, ESG 101

digfin group, ESG now a third of MioTech’s A.I. business,  September 9, 2019.

MioTech builds A.I.-based solutions to help buy sides get insight from data analytics, and to help sell-side research departments and private banks’ relationship managers tell data-driven stories. It bases its service on building a library of cross-references (a “knowledge graph”) around a multitude of data points on Asian companies. The idea is to use big-data correlations to spot patterns....

Recyclers turn to AI robots after waste import bans

Adam Green, Recyclers turn to AI robots after waste import bans, Financial Times, 2020/7/1.

To recycle in a cost-effective, comprehensive and safe way, goods must be broken down into their constituent commodities to be sold on, in a process that has been likened to “unscrambling an egg”....

Automation often stokes anxiety about mass unemployment, but the recycling sector has been struggling to find enough workers. The US waste and recycling industry has suffered labour shortages in recent years. By limiting the influx of foreign workers to do jobs locals are not keen on, the UK’s departure from the EU is expected to hit the UK’s waste management sector hard. 

“This technology is creating a sustainable workforce for jobs that aren’t being filled,” says Mr Wirth. “These are the dull, dirty, dangerous kind of jobs which robotics and AI is perfect for.”

Best of Bach - Classical Guitar Compilation - BWV

Best of Bach - Classical Guitar Compilation - BWV, 2019年4月15日

Nancy Duarte uncovers common structure of greatest communicators

TEDxEast - Nancy Duarte uncovers common structure of greatest communicators, 11/11/2010

 At 17:50:

You can change your life. You can change the world that you have control over, you can change your sphere. 

毛佩琦譯，簡報女王的故事力！矽谷最有說服力的不敗簡報聖經，商業周刊，2020
Nancy Duarte, Resonate: Present Visual Stories that Transform Audiences, Wiley, 2010.

3Blue1Brown 驚人的視覺化數學影片

But what *is* a Neural Network? | Chapter 1, deep learning 

(主頁面，所有的主題都有好幾部影片)

劉士堉吉他

guitarist641

Classical guitar by Ana Vidovic

ANA VIDOVIC - LIVE CONCERT - LAMBRECHT – CLASSICAL GUITAR EVENTS

首播日期：2019年7月8日

學生時代彈過幾年的古典吉他。

Advanced Deep Learning and Reinforcement Learning by DeepMind x UCL

DeepMind x UCL, Deep Learning Lectures (12 lectures), 2020/6/1

DeepMind x UCL, Advanced Deep Learning and Reinforcement Learning (18 lecture), GitHub and YouTube, 2018

Optimal Transport

Professor Cédric Villani, “Optimal Transport Theory”, New Frontiers in Mathematics, 2018/1/17.

Making Better Fulfillment Decisions on the Fly in an Online Retail Environment

Jason Acimovic and Stephen C. Graves, Making Better Fulfillment Decisions on the Fly in an Online Retail Environment, Manufacturing & Service Operations Management, Volume 17, Issue 1, Winter 2015. (2017 M&SOM Best Paper Award)

We develop a heuristic that makes fulfillment decisions by minimizing the immediate outbound shipping cost plus an estimate of future expected outbound shipping costs (*). These estimates are derived from the dual values of a transportation linear program (LP). In our experiments on industry data, we capture 36% of the opportunity gap assuming clairvoyance, leading to reductions in outbound shipping costs on the order of 1%. These cost savings are achieved without any deterioration in customer service levels or any increase in holding costs.

東北角偏鄉小校 雲端教育大爆發

商業周刊，東北角偏鄉小校 雲端教育大爆發，2020年4月24日

Chip Placement with Deep Reinforcement Learning

Azalia Mirhoseini, et al., Chip Placement with Deep Reinforcement Learning, arXiv:2004.10746.

In this work, we present a learning-based approach to chip placement, one of the most complex and time-consuming stages of the chip design process. Unlike prior methods, our approach has the ability to learn from past experience and improve over time. In particular, as we train over a greater number of chip blocks, our method becomes better at rapidly generating optimized placements for previously unseen chip blocks. To achieve these results, we pose placement as a Reinforcement Learning (RL) problem and train an agent to place the nodes of a chip netlist onto a chip canvas. To enable our RL policy to generalize to unseen blocks, we ground representation learning in the supervised task of predicting placement quality. By designing a neural architecture that can accurately predict reward across a wide variety of netlists and their placements, we are able to generate rich feature embeddings of the input netlists. We then use this architecture as the encoder of our policy and value networks to enable transfer learning. Our objective is to minimize PPA (power, performance, and area), and we show that, in under 6 hours, our method can generate placements that are superhuman or comparable on modern accelerator netlists, whereas existing baselines require human experts in the loop and take several weeks.

How to motivate yourself to change your behavior

Tali Sharot, How to motivate yourself to change your behavior, TEDxCambridge, 2014/10/28

How to do it in 3 steps: social incentives, immediate reward, and progress monitoring (at 15:22)

COS116: The Computational Universe

Sanjeev Arora, Computer Science 116 The Computational Universe, Princeton University, 2006.

幾年前，我教資管系一下的網際網路應用。心想計算機概論已經教通訊軟體的操作，所以網際網路應用改教重要的概念，取材來自幾本書，並簡化內容，例如資訊經營法則、Networked Life、Mining of Massive Datasets。當時有許多學生棄選，不知道是不是英文字 (部分有中譯) 太多？還是概念比較抽象？方程式太多？ 

Prof. Arora 這門課的授課對象是任何學系的學生，可以視為台灣的通識教育。最近使用 Amazon Kindle (電子書閱讀器) 閱讀其講義，發現講得更生活化，例如以圖書館員取書的例子，說明計算機結構中快取 (cache) 的概念。再一次呼應了我之前的說法，世界級的研究型大學，教學也是令人驚艷。

另外一點，美國的學費驚人，而且有廣大的社會捐獻和好的投資績效，所以經費充足、課程的支援較好。這門課有實驗，大都是助教負責寫講義和上課。但是，台灣的大學學費凍漲，自然無法和其相比。 

給青年知識追求者的十封信

朱敬一，給青年知識追求者的十封信，聯經出版，2018

經過十餘年的沉潛，走在求知路上的朱敬一院士，始終心繫台灣的青年學子。 

為什麼要閱讀經典

王汎森，為什麼要閱讀經典？，風傳媒，2019-11-16  

林肯正式的教育背景很淺，可是他的文章和演講卻極為動人，我認為他主要的精神及思想資源是不斷地精讀包括《聖經》在內的經典。這些經典深入他的精神層次，除了源源不斷地提供許多精彩的養分之外，也形成了一個規模、一個框架、一種內在的雄渾的氣氛與節奏，再加上他在適當的時候套引經典，使得他的演講及文字非常動人。...

外交官的涵養與應變

朱敬一，外交官的涵養與應變，風傳媒，2020-04-30

什麼是「外交」，不需要我來定義；維基百科裡對於外交的歷史緣起、功能特色、正式與非正式外交、外交豁免等，都有詳盡的介紹。但是「外交是什麼」與「如何做好外交」，是兩個不同層面的問題。我是經濟學研究者，算是熟悉國際經濟體系與經濟學理論；我做過科技首長、中央研究院副院長，了解科技發展與產業，也對台灣官僚體系的運作了然於心。但是如何將經濟學與科技產業的知識背景「融入」外交場域，卻不是一件容易的事。這裡可以先說結論：我認為做好外交官，只有四個字，涵養、應變。仔細一點分析，應變又來自於涵養。所以外交官的本事，就是以「涵養」二字為根基。

A Human-Centered Evaluation of a Deep Learning System Deployed in Clinics for the Detection of Diabetic Retinopathy

Emma Beede, Elizabeth Baylor, Fred Hersch, Anna Iurchenko, Lauren Wilcox, Paisan Ruamviboonsuk, Laura M. Vardoulakis, A Human-Centered Evaluation of a Deep Learning System Deployed in Clinics for the Detection of Diabetic Retinopathy, CHI '20: Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, April 2020, Pages 1–12, https://doi.org/10.1145/3313831.3376718.

Referral Determinations

All images were initially assessed by a nurse then sent to an ophthalmologist for review. The ability to assess fundus photos for DR varied from nurse to nurse. While most nurses told us they felt comfortable assessing for the presence of DR, they didn’t know how to determine the severity if present. P4 told us, “I know if it’s not normal, but I don’t know what to call it.” To make the ultimate decision of whether a patient needs to be referred to an ophthalmologist for an exam and potentially for treatment, the nurse turned to the ophthalmologist or retinal specialist, who are most often remote.

NBDT: Neural-Backed Decision Trees

Alvin Wan, Lisa Dunlap, Daniel Ho, Jihan Yin, Scott Lee, Henry Jin, Suzanne Petryk, Sarah Adel Bargal, Joseph E. Gonzalez, NBDT: Neural-Backed Decision Trees,  arXiv:2004.00221, 2020

We forgo this dilemma by creating Neural-Backed Decision Trees (NBDTs) that (1) achieve neural network accuracy and (2) require no architectural changes to a neural network. NBDTs achieve accuracy within 1% of the base neural network on CIFAR10, CIFAR100, TinyImageNet, using recently state-of-the-art WideResNet; and within 2% of EfficientNet on ImageNet. This yields state-of-the-art explainable models on ImageNet, with NBDTs improving the baseline by ~14% to 75.30% top-1 accuracy. Furthermore, we show interpretability of our model's decisions both qualitatively and quantitatively via a semi-automatic process. Code and pretrained NBDTs can be found at this https URL.

Digital Social Innovation (數位社會創新)

Audrey Tang (唐鳳), Digital Social Innovation (數位社會創新), 2020/4/21

Let Taiwan into the World Health Organisation

Economist, Let Taiwan into the World Health Organisation, 2020/03/26.

Spare a moment and admire Taiwan. Its handling of the new coronavirus pandemic has so far saved many, many lives. The figures tell the story. A country of 24m, it has far fewer infections than its neighbours: just 235 as of March 25th, with only two deaths...

原則：生活和工作 (Principles: Life and Work)

陳世杰、諶悠文、戴至中譯，原則：生活和工作，商業周刊，2018
Ray Dalio, Principles: Life and Work, Simon & Schuster, 2017 (excerpt)

 瑞．達利歐出身美國普通中產家庭，26歲時被投資公司炒魷魚，在自己的兩房公寓室白手起家創辦了橋水，並在接下來超過42年裡，把橋水打造成了獲《財星》（Fortune）雜誌評選為美國第五重要的私人公司。現在橋水管理資金超過1,500億美元，截至2015年年底，盈利超過450億美元。達利歐曾成功預測2008年金融危機，成為華爾街教父級大神。 

一路以來，達利歐曾入選世界百大最具影響力人物［《時代》（Time）］與百大富豪［《富比世》（Forbes）］，而且由於他獨特的投資原則改變了業界，《CIO》更稱他是「投資界的史蒂夫．賈伯斯」。 

他是怎麼辦到的？靠的是「原則」！他從1982年看錯墨西哥債務危機、狠狠跌交的經驗中吸取教訓，提煉決策標準，日積月累，總結成一組「原則」，包含21條高層原則、139條中原則和365條分原則，涵蓋為人處事、公司管理兩大方面，是橋水的員工手冊，橋水依循進行日常管理，也是橋水成為全球最強避險基金的祕密。

光學檢測獲利王牧德科技

吳靜芳，光學檢測獲利王！高CP值媲美世界龍頭，百大企業九成是牧德客戶，天下雜誌 672期，2019-05-06

「現在 PCB 愈做愈小，朝30微米以下發展，人工目檢漸漸不可行，」工業技術研究院產科國際所分析師黃仲宏觀察，在自動化趨勢下，AOI 早已是PCB廠的剛性需求。...

財務自由的人生

楊應超，財務自由的人生：跟著首席分析師楊應超學華爾街的投資技巧和工作效率，40 歲就過 FIRE，天下文化，2019

機構投資者》、《金融時報》、《格林威治》、《亞元雜誌》評選為第一名首席外資分析師楊應超不藏私，首度公開他征戰全球頂尖投行的投資和工作心法，幫助你40歲前達到FIRE，過不再為錢煩惱的優質人生！