(年輕人) 如何脫離貧窮

一般人都可以經由努力而達到的方法 (註 12) 

找工作可分成學校教育、個人努力、企業、和社會大環境等四大面向。不論是 22 K 政策、學費、企業的在職訓練等等，都不是我們可以掌握的。當大環境不容易改變，個人只有自求多福。所以，本文希望能夠提供一些可行的辦法，以幫助年輕人。方法是在大學四年 (和終身學習) (註 1)，培養三大類的能力，一是一般的能力，語言、電腦使用，二是專業，三是強化個人特質和團隊合作精神。有了這些能力，也才有機會選擇自己喜歡的工作。 

Interpretable predictive maintenance for hard drives

Maxime Amram, Jack Dunn, Jeremy J.Toledano, and Ying Daisy Zhuo, Interpretable predictive maintenance for hard drives, Machine Learning with Applications, Volume 5, 15 September 2021, 100042.

Existing machine learning approaches for data-driven predictive maintenance are usually black boxes that claim high predictive power yet cannot be understood by humans. This limits the ability of humans to use these models to derive insights and understanding of the underlying failure mechanisms, and also limits the degree of confidence that can be placed in such a system to perform well on future data. We consider the task of predicting hard drive failure in a data center using recent algorithms for interpretable machine learning. We demonstrate that these methods provide meaningful insights about short- and long-term drive health, while also maintaining high predictive performance. We also show that these analyses still deliver useful insights even when limited historical data is available, enabling their use in situations where data collection has only recently begun.

Deep Learning for AI

Yoshua Bengio, Yann Lecun, and Geoffrey Hinton, Deep Learning for AI, Communications of the ACM, July 2021, Vol. 64 No. 7, Pages 58-65.

Research on artificial neural networks was motivated by the observation that human intelligence emerges from highly parallel networks of relatively simple, non-linear neurons that learn by adjusting the strengths of their connections. This observation leads to a central computational question: How is it possible for networks of this general kind to learn the complicated internal representations that are required for difficult tasks such as recognizing objects or understanding language? Deep learning seeks to answer this question by using many layers of activity vectors as representations and learning the connection strengths that give rise to these vectors by following the stochastic gradient of an objective function that measures how well the network is performing. It is very surprising that such a conceptually simple approach has proved to be so effective when applied to large training sets using huge amounts of computation and it appears that a key ingredient is depth: shallow networks simply do not work as well.

Boeing's 737 MAX: A Failure of Management, Not Just Technology

Michael A. Cusumano, Boeing's 737 MAX: A Failure of Management, Not Just Technology, Communications of the ACM, January 2021, Vol. 64, No. 1, Pages 22-25.

The congressional report had extensive access to company email and documents as well as detailed media coverage. These sources all describe the same decisions along with gradual but fundamental changes in Boeing's strategy and culture.

The Future of Supply Chains

Paul Marks, The Future of Supply Chains, Communications of the ACM, July 2021, Vol. 64, No. 7, Pages 19-21.

Today's supply chains are labor-intensive and expensive to run. A number of autonomous systems that reduce the human factor are about to change all that.

What do the sidewalks around us, the airspace above us, interstate freeways, and deep ocean shipping lanes have in common? The answer is that they are all places where developers of autonomous technology are trying to revolutionize the economics of supply chains. The plan is to use robotic technology to deliver anything from packages to take-out food, groceries, or bulk freight in ways that can reduce the logistics industry's dependence on that most expensive of supply chain costs: human labor. if the use of electric drivetrains can cut carbon emissions too, so much the better.

Revenue Management and Pricing Analytics

Guillermo Gallego and Huseyin Topaloglu, Revenue Management and Pricing Analytics, Springer, 2019.

The book is divided into three parts: traditional revenue management, revenue management under customer choice, and pricing analytics. Each part is approximately of the same length and written in a self-contained way, so readers can read them independently, although reading the first part may make the second part easier to understand. Each chapter ends with bibliographical notes where the reader can find the sources of the material covered as well as many useful references. Proofs of some important technical results can be found in the appendix of each chapter. Solving the end-of-chapter problems helps reinforce the material in the book, with some of the questions expanding on the subject....

There is enough material in the book for a full-semester course for advanced undergraduate or master’s students. Parts I and II can be covered in about 9 weeks and Part III in about 4weeks excluding the last two chapters on online learning and competition, which can be assigned as independent readings. 

Thinking, Fast and Slow (快思慢想)

Daniel Kahneman, Thinking, Fast and Slow, Penguin Group UK, 2012

作者融合數十年 (心理) 學界的研究，說明人是如何思考和選擇，遣詞用字清楚易懂，值得花時間細讀 (註 1)。

作者和 Amos Tversky 提出展望理論 (Prospect theory)，比經濟學的預期效用理論 (expected utility theory) 更精準地描述人類的決策方法，對行為經濟學有深遠的影響，因而得到 2002 年的諾貝爾經濟學獎 (註 2)。

至於在企業的應用，可以參考作者的另外一篇文章。

(21/6/17) 有朋友在群組，轉貼台南市發布的「零接觸採檢站」新聞稿影片。看到整個流程的設計，覺得很有教育意義，所以就轉發到其他的群組。有趣的是，有人覺得這個是「台南市政府的大內宣」。我就上網查了一下資料，實際的狀況是，台積電要捐八部車子，這只是第一部，而且有許多媒體的報導。快思慢想第七章 (A machine for jumping to conclusions) 就是在講這種人類思維的現象。 

(註 1)  英文版的內文沒有說明或任何記號，所以不容易知道文章何處有註釋。中譯本的內文有譯注；但是，刪除 51 頁的註釋和索引；例如出現多次的 WYSIATI (what you see is all there is) 目錄也沒有，除非第一次就讀到且記住，否則很難瞭解。

(註 2) 原文 Daniel Kahneman and Amos Tversky, Prospect Theory: An Analysis of Decision Under Risk, Econometrica, Vol. 47, No. 2, 1979, pp. 263-292。Amos Tversky 於 1996 過世，所以沒有得獎。書中 279 頁說明其成功的原因

Knowledge of perception and ignorance about decision theory both contributed to a large step forward in our research.

Analytics and Bikes: Riding Tandem with Motivate to Improve Mobility

Daniel Freund, S. G. Henderson, E. O’Mahony, and D. B. Shmoys, Analytics and Bikes: Riding Tandem with Motivate to Improve Mobility, INFORMS Journal on Applied Analytics, Vol. 49, No. 5, 2019, Pages 307-396. (First Prize: Wagner Prize for Excellence in Operations Research Practice)

Bike-sharing systems are now ubiquitous across the United States. We have worked with Motivate, the operator of the systems in, for example, New York, Chicago, and San Francisco, to both innovate a data-driven approach to managing their day-to-day operations and provide insight on several central issues in the design of its systems. This work required the development of a number of new optimization models, characterization of their mathematical structure, and use of this insight in designing algorithms to solve them. Here, we focus on two particularly high-impact projects: an initiative to improve the allocation of docks to stations and the creation of an incentive scheme to crowdsource rebalancing. Both of these projects have been fully implemented to improve the performance of Motivate’s systems across the country; for example, the Bike Angels program in New York City yields a system-wide improvement comparable with that obtained through Motivate’s traditional rebalancing efforts at far less financial and environmental cost.

Reinforcement learning is enough to reach general AI

David Silver, Satinder Singh, Doina Precup, and Richard S. Sutton, Reward is enough, Artificial Intelligence, Volume 299, October 2021, 103535.

In this article we hypothesise that intelligence, and its associated abilities, can be understood as subserving the maximisation of reward. Accordingly, reward is enough to drive behaviour that exhibits abilities studied in natural and artificial intelligence, including knowledge, learning, perception, social intelligence, language, generalisation and imitation. This is in contrast to the view that specialised problem formulations are needed for each ability, based on other signals or objectives. Furthermore, we suggest that agents that learn through trial and error experience to maximise reward could learn behaviour that exhibits most if not all of these abilities, and therefore that powerful reinforcement learning agents could constitute a solution to artificial general intelligence.

Supply chain of COVID vaccines

Asher Mullard, How COVID vaccines are being divvied up around the world: Canada leads the pack in terms of doses secured per capita, Nature,  30 November 2020.

But the makers of the three vaccines that seem closest to widespread distribution — AstraZeneca, Pfizer and Moderna — estimate a total production capacity of 5.3 billion doses for 2021, which could cover between 2.6 billion and 3.1 billion people, depending on whether AstraZeneca’s vaccine is administered in two doses or one and a half (see ‘Vaccine pre-orders’).

It's Okay.

Golden Buzzer: Nightbirde's Original Song Makes Simon Cowell Emotional - America's Got Talent 2021.

Nightbirde, “You can’t wait until life isn’t hard anymore before you decide to be happy.” 

台灣的疫苗戰略新思維

黃韻如、賴育宏、鄭如韻，台灣作為防疫優等生　為何在「疫苗戰」失去主控權？（上），ETtoday新聞雲，2021年02月09日

近日，各界輿論聚焦在疫苗採購中僵化的《政府採購法》，關切中央流行疫情指揮中心在國際疫苗的採購談判及取得進度、國產疫苗供應鏈的開發瓶頸等問題時，本評論希望拉高討論的視角，以國家資本與國際人脈的戰略角度，參考新加坡與以色列經驗，審視台灣在這場世紀抗疫之戰，為什麼在疫苗賽事之中落後的根本原因，也藉此窺探台灣在生醫新創的國際產業生態圈所面臨到的困境。

黃韻如、鄭如韻、高子翔、蘇育平， 以色列疫苗採購奪冠，為何台灣不行？台灣生技人才出走的警訊，ETtoday新聞雲，2021年02月27日 

本文分析重點有二：

一、以人口不到千萬的年輕國家以色列為例，如何從不可抗拒的大離散（diaspora）動盪飄搖命運中遍地開花，在緊繃的疫情中，透過總理納坦雅胡與 Pfizer 猶太裔執行長Albert Bourla的 17 通電話，讓最初疫情失控的以色列，優先取得疫苗及施打保障 [5]，扭轉局面。

二、帶領讀者認識台灣生技人才同樣經歷出走全球的離散、遍地開花，進而組織自發性的台灣生技社群，期許台灣政府正視並鏈接散居在海外的台灣生技製藥菁英。

台灣在疫苗採購上面臨許多不可抗力的因素，文末，本團隊建議台灣，如何藉由防疫黃金三角的架構，善用台灣人才、鏈結國際人脈網絡、增強國內產業體質，優化深化台灣「平時備戰，戰時作戰」的韌性與永續。

A Lyapunov Analysis of Accelerated Methods in Optimization

Ashia C. Wilson, Ben Recht, and Michael I. Jordan, A Lyapunov Analysis of Accelerated Methods in Optimization, Journal of Machine Learning Research, 2021, Vol. 22, No. 113, pp. 1−34.

Accelerated optimization methods, such as Nesterov’s accelerated gradient method, play a significant role in optimization. Several accelerated methods are provably optimal under standard oracle models. Such optimality results are obtained using a technique known as estimate sequences which yields upper bounds on convergence properties. The technique of estimate sequences has long been considered difficult to understand and deploy, leading many researchers to generate alternative, more intuitive methods and analyses. We show there is an equivalence between the technique of estimate sequences and a family of Lyapunov functions in both continuous and discrete time. This connection allows us to develop a unified analysis of many existing accelerated algorithms, introduce new algorithms, and strengthen the connection between accelerated algorithms and continuous-time dynamical systems.

Convex Clustering: Model, Theoretical Guarantee and Efficient Algorithm

Defeng Sun, Kim-Chuan Toh, and Yancheng Yuan, Convex Clustering: Model, Theoretical Guarantee and Efficient Algorithm, Journal of Machine Learning Research, 2021, Vol. 22, No. 9, pp. 1−32.

Clustering is a fundamental problem in unsupervised learning. Popular methods like Kmeans, may suffer from poor performance as they are prone to get stuck in its local minima. Recently, the sum-of-norms (SON) model (also known as the convex clustering model) has been proposed by Pelckmans et al. (2005), Lindsten et al. (2011) and Hocking et al. (2011). The perfect recovery properties of the convex clustering model with uniformly weighted all-pairwise-differences regularization have been proved by Zhu et al. (2014) and Panahi et al. (2017). However, no theoretical guarantee has been established for the general weighted convex clustering model, where better empirical results have been observed. In the numerical optimization aspect, although algorithms like the alternating direction method of multipliers (ADMM) and the alternating minimization algorithm (AMA) have been proposed to solve the convex clustering model (Chi and Lange, 2015), it still remains very challenging to solve large-scale problems. In this paper, we establish sufficient conditions for the perfect recovery guarantee of the general weighted convex clustering model, which include and improve existing theoretical results in (Zhu et al., 2014; Panahi et al., 2017) as special cases. In addition, we develop a semismooth Newton based augmented Lagrangian method for solving large-scale convex clustering problems. Extensive numerical experiments on both simulated and real data demonstrate that our algorithm is highly efficient and robust for solving large-scale problems. Moreover, the numerical results also show the superior performance and scalability of our algorithm comparing to the existing first-order methods. In particular, our algorithm is able to solve a convex clustering problem with 200,000 points in R^3 in about 6 minutes.

The Nutritious Supply Chain: Optimizing Humanitarian Food Assistance

Koen Peters, Sérgio Silva, Rui Gonçalves, Mirjana Kavelj, Hein Fleuren, Dick den Hertog, Ozlem Ergun, and Mallory Freeman, The Nutritious Supply Chain: Optimizing Humanitarian Food Assistance, INFORMS Journal on Optimization, Volume 3, Issue 2, Spring 2021, Pages:200–226.

The World Food Programme (WFP) is the largest humanitarian agency fighting hunger worldwide, reaching approximately 90 million people with food assistance across 80 countries each year. To deal with the operational complexities inherent in its mandate, WFP has been developing tools to assist its decision makers with integrating supply chain decisions across departments and functional areas. This paper describes a mixed integer linear programming model that simultaneously optimizes the food basket to be delivered, the sourcing plan, the delivery plan, and the transfer modality of a long-term recovery operation for each month in a predefined time horizon. By connecting traditional supply chain elements to nutritional objectives, we are able to make significant breakthroughs in the operational excellence of WFP’s most complex operations. We show three examples of how the optimization model is used to support operations: (1) to reduce the operational costs in Iraq by 12% without compromising the nutritional value supplied, (2) to manage the scaling-up of the Yemen operation from three to six million beneficiaries, and (3) to identify sourcing strategies during the El Niño drought of 2016.

台灣的疫情和疫苗

 臉友轉貼。我竟然耐心地看完了，順便追蹤原作者。文章裡面介紹的兩本書，列為待讀清單。

邱斯華，6月4日上午6:44 

<我對疫苗的想法>

最近在 Line 群組裡老是有群友在黨同伐異、互添仇恨值，搞的我只剩下退群求個清靜，或是寫下我自己的分析看法，讓雙方冷靜停戰這兩個選擇。不敢說我看的一定正確，也祈請各方指正我的偏誤。

各位熟知我文章風格的朋友都知道我是個中立派，因為在我眼中，在歷史上也是，沒有永遠真善美的黨派。在瞬變的世局中，我們只能盡量摘掉有色眼鏡來看清這個真實世界，儘量逼近真實。

Probabilistic Machine Learning

 "Probabilistic Machine Learning" - a book series by Kevin Murphy

Kevin Patrick Murphy, Probabilistic Machine Learning: An Introduction, MIT Press, 2021. (Python codes)

新的版本，作者花了很多時間整理。以第一部分的數學基礎為例，使用機器學習說明相關的概念。也納入時事，例如 2.3.1 節的 Testing for COVID-19。 

台積電捐贈「零接觸採檢站」予奇美醫院 工研院「正壓採檢亭」詢問度高

公視晚間新聞，台積電捐贈「零接觸採檢站」予奇美醫院 工研院「正壓採檢亭」詢問度高，2021年5月28日

Reducing Conservatism in Robust Optimization

E. Roos and D. den Hertog, Reducing conservatism in robust optimization, INFORMS Journal on Computing, Fall 2020, Volume 32, Issue 4, pp. 1109–1127.

Although robust optimization is a powerful technique in dealing with uncertainty in optimization, its solutions can be too conservative. More specifically, it can lead to an objective value much worse than the nominal solution or even to infeasibility of the robust problem. In practice, this can lead to robust solutions being disregarded in favor of the nominal solution. This conservatism is caused by both the constraint-wise approach of robust optimization and its core assumption that all constraints are hard for all scenarios in the uncertainty set. This paper seeks to alleviate this conservatism by proposing an alternative robust formulation that condenses all uncertainty into a single constraint, binding the worst-case expected violation in the original constraints from above. Using recent results in distributionally robust optimization, the proposed formulation is shown to be tractable for both right- and left-hand side uncertainty. A computational study is performed with problems from the NETLIB library. For some problems, the percentage of uncertainty is magnified fourfold in terms of increase in objective value of the standard robust solution compared with the nominal solution, whereas we find solutions that safeguard against over half the violation at only a 10th of the cost in objective value. For problems with an infeasible standard robust counterpart, the suggested approach is still applicable and finds both solutions that safeguard against most of the uncertainty at a low price in terms of objective value.