預備研究生 (4 + 1) (包含資料科學相關研究所和職涯準備)

預備研究生規定 (資訊處，新增加同學的問題)

• 培養
• 商業分析實驗室 (Business Analytics Laboratory)：論文。
• 人少，老師 (每週) 單獨討論。可於辦公室時間詢問庭煜。
• 學生的光，老師看得見。

Customer Choice Models vs. Machine Learning

Jacob Feldman, Dennis J. Zhang, Xiaofei Liu, and Nannan Zhang (2021) Customer Choice Models vs. Machine Learning: Finding Optimal Product Displays on Alibaba. Operations Research 70(1):309-328. (Best OM Paper in Operations Research Award: Finalist, pdf, implementation details)

Optimization in Online Content Recommendation Services

Omar Besbes, Yonatan Gur, Assaf Zeevi, Optimization in Online Content Recommendation Services: Beyond Click-Through Rates, 18(1), pp. 15–33, Manufacturing & Service Operations Management, Volume 18, Issue 1, Winter 2016. 

A new class of online services allows Internet media sites to direct users from articles they are currently reading to other content they may be interested in. This process creates a “browsing path” along which there is potential for repeated interaction between the user and the provider, giving rise to a dynamic optimization problem. A key metric that often underlies this recommendation process is the click-through rate (CTR) of candidate articles. Whereas CTR is a measure of instantaneous click likelihood, we analyze the performance improvement that one may achieve by some lookahead that accounts for the potential future path of users. To that end, by using some data of user path history at major media sites, we introduce and derive a representation of content along two key dimensions: clickability, the likelihood to click to an article when it is recommended; and engageability, the likelihood to click from an article when it hosts a recommendation. We then propose a class of heuristics that leverage both clickability and engageability, and provide theoretical support for favoring such path-focused heuristics over myopic heuristics that focus only on clickability (no lookahead). We conduct a live pilot experiment that measures the performance of a practical proxy of our proposed class, when integrated into the operating system of a worldwide leading provider of content recommendations, allowing us to estimate the aggregate improvement in clicks per visit relative to the CTR-driven current practice. The documented improvement highlights the importance and the practicality of efficiently incorporating the future path of users in real time.

Machine learning and statistical testing at vantagescore

Kimberly Adams, How one credit scoring company is thinking about financial inclusion, Marketplace Tech,  JULY 8, 2022. (more around July 2022)

作業研究的持續改善

下學年要教我喜歡的作業研究。依照往例，再一次更新之前的版本。

第一學期，考慮確定的環境和有限資源下，使用最佳化 (optimization)，以最大化 (例如利潤) 或最小化目標  (例如成本)。第二學期，準備教不確定性環境下，如何分析、最佳化和做決策 (控制)。

Online Network Revenue Management Using Thompson Sampling

Kris Johnson Ferreira, David Simchi-Levi, and He Wang. (2018). “Online network revenue management using Thompson sampling.” Operations Research, 66(6), 1586-1602. (Supplemental Material, code, MSOM Society 2021 Operations Research Best OM Paper Award)

We consider a price-based network revenue management problem in which a retailer aims to maximize revenue from multiple products with limited inventory over a finite selling season. As is common in practice, we assume the demand function contains unknown parameters that must be learned from sales data. In the presence of these unknown demand parameters, the retailer faces a trade-off commonly referred to as the “exploration-exploitation trade-off.” Toward the beginning of the selling season, the retailer may offer several different prices to try to learn demand at each price (“exploration” objective). Over time, the retailer can use this knowledge to set a price that maximizes revenue throughout the remainder of the selling season (“exploitation” objective). We propose a class of dynamic pricing algorithms that builds on the simple, yet powerful, machine learning technique known as “Thompson sampling” to address the challenge of balancing the exploration-exploitation trade-off under the presence of inventory constraints. Our algorithms have both strong theoretical performance guarantees and promising numerical performance results when compared with other algorithms developed for similar settings. Moreover, we show how our algorithms can be extended for use in general multiarmed bandit problems with resource constraints as well as in applications in other revenue management settings and beyond.

為什麼工 (資) 管的課程看起來很雜？

(2022) 因為橫跨兩個學院 (工和商)，工業系龐雜的內容，和資管系一樣，所以修改一下標題。其實，我的部落格如何選填大學志願，就是以四個專業和李國鼎的話 (第 17 頁)，說明這兩個系的重要性。

(2014) 常常聽到學生有這樣的疑惑，我試著以營收管理說明之；針對固定 (如旅館房間、網頁上廣告空間) 且易過時 (如機位、時裝) 的容量或庫存供給下，如何有效地分配庫存 (屬於生管) 和 (動態) 定價 (屬於行銷)，以最大化企業之營收；詳細的內容可以參考我的課程。

7 real-world applications of reinforcement learning

 Joy Zhang, 7 real-world applications of reinforcement learning, gocoder, February 17, 2022

1. Autonomous driving with Wayve

2. Personalizing your Netflix recommendations

3. Optimizing inventory levels for Walmart

4. Improving search engine results with search.io

5. Improving language models with OpenAI's WebGPT

6. Trading on the financial markets with IBM's DSX platform

7. Robotics with the University of California, Berkeley

台積電的訂價策略

陳良榕，台積電怎麼贏？， 天下雜誌 609 期，2016-10-25

經營人的學習與成長

張忠謀，經營人的學習與成長，台灣玉山科技協會20周年慶祝大會暨論壇-晚宴，10月26日

經營人的學習與成長

張忠謀，經營人的學習與成長，台灣玉山科技協會20周年慶祝大會暨論壇-晚宴，2021年10月26日

Optimization modelling: A practical approach

Ruhul Amin Sarker and Charles S. Newton, Optimization modelling: A practical approach, CRC Press, 2007. 

Although a useful and important tool, the potential of mathematical modelling for decision making is often neglected. Considered an art by many and weird science by some, modelling is not as widely appreciated in problem solving and decision making as perhaps it should be. And although many operations research, management science, and optimization books touch on modelling techniques, the short shrift they usually get in coverage is reflected in their minimal application to problems in the real world. Illustrating the important influence of modelling on the decision making process, Optimization Modelling: A Practical Approach helps you come to grips with a wide range of modelling techniques.

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. 

Targeting Prospective Customers: Robustness of Machine-Learning Methods to Typical Data Challenges

Duncan Simester, Artem Timoshenko, and Spyros I. Zoumpoulis, Targeting Prospective Customers: Robustness of Machine-Learning Methods to Typical Data Challenges, Management Science, 2020, Vol. 66, No. 6, Pages: 2495–2522.

We investigate how firms can use the results of field experiments to optimize the targeting of promotions when prospecting for new customers. We evaluate seven widely used machine-learning methods using a series of two large-scale field experiments. The first field experiment generates a common pool of training data for each of the seven methods. We then validate the seven optimized policies provided by each method together with uniform benchmark policies in a second field experiment. The findings not only compare the performance of the targeting methods, but also demonstrate how well the methods address common data challenges. Our results reveal that when the training data are ideal, model-driven methods perform better than distance-driven methods and classification methods. However, the performance advantage vanishes in the presence of challenges that affect the quality of the training data, including the extent to which the training data captures details of the implementation setting. The challenges we study are covariate shift, concept shift, information loss through aggregation, and imbalanced data. Intuitively, the model-driven methods make better use of the information available in the training data, but the performance of these methods is more sensitive to deterioration in the quality of this information. The classification methods we tested performed relatively poorly. We explain the poor performance of the classification methods in our setting and describe how the performance of these methods could be improved.

Efficient Large-Scale Internet Media Selection Optimization for Online Display Advertising

Courtney Paulson, Lan Luo, and Gareth M. James, “Efficient Large-ScaleInternet Media Selection Optimization for Online Display Advertising,” Journal of Marketing Research, vol. 55, 2018, pp. 489-506. 

In today's digital market, the number of websites available for advertising has ballooned into the millions. Consequently, firms often turn to ad agencies and demand-side platforms (DSPs) to decide how to allocate their Internet display advertising budgets. Nevertheless, most extant DSP algorithms are rule-based and strictly proprietary. This article is among the first efforts in marketing to develop a nonproprietary algorithm for optimal budget allocation of Internet display ads within the context of programmatic advertising. Unlike many DSP algorithms that treat each ad impression independently, this method explicitly accounts for viewership correlations across websites. Consequently, campaign managers can make optimal bidding decisions over the entire set of advertising opportunities. More importantly, they can avoid overbidding for impressions from high-cost publishers, unless such sites reach an otherwise unreachable audience. The proposed method can also be used as a budget-setting tool, because it readily provides optimal bidding guidelines for a range of campaign budgets. Finally, this method can accommodate several practical considerations including consumer targeting, target frequency of ad exposure, and mandatory media coverage to matched content websites.

Algorithm: Coordinate descent algorithm for budget optimization problem (7).

Dataflow-as-a-Service of SambaNova

Inside AI, 2021/4/14

SambaNova, a developer of AI hardware and software systems, raised $676M in financing that values the startup at over $5.1B. The Nvidia competitor makes chips for AI processes, which it uses to build servers and AI software that it leases to other businesses.

Charting a business course for reinforcement learning

Jacomo Corbo, Oliver Fleming, and Nicolas Hohn, It’s time for businesses to chart a course for reinforcement learning, McKinsey, April 1, 2021.

Broadly speaking, we see reinforcement learning delivering this value across the business, with potential applications in every business domain and industry (Exhibit 2). Some of the near-term applications for reinforcement learning fall into three categories: speeding design and product development, optimizing complex operations, and guiding customer interactions.

 Exhibit 2 some applications.

To be sure, implementing reinforcement learning is a challenging technical pursuit. A successful reinforcement learning system today requires, in simple terms, three ingredients:

1. A well-designed learning algorithm with a reward function. 
2. A learning environment.  
3. Compute power. 

Computing power is better than compute power. 

Feedback Control in Programmatic Advertising

N. Karlsson, Feedback Control in Programmatic Advertising: The Frontier of Optimization in Real-Time Bidding, IEEE Control Systems Magazine, vol. 40, no. 5, pp. 40-77, Oct. 2020.

Feedback control is critical in the scalable optimization of Internet advertising, and it is, therefore, an enabling technology. However, it is challenging to model the plant and design the controller because the plant is nonlinear, time varying, stochastic, and poorly known. A closed-loop system model easily becomes unrealistic or extremely complicated and intractable to analyze.

How we use AutoML, Multi-task learning and Multi-tower models for Pinterest Ads

Ernest Wang, How we use AutoML, Multi-task learning and Multi-tower models for Pinterest Ads, Aug 20, 2020.

People come to Pinterest in an exploration mindset, often engaging with ads the same way they do with organic Pins. Within ads our mission is to help Pinners go from inspiration to action by introducing them to the compelling products and services that advertisers have to offer. A core component of the ads marketplace is predicting engagement of Pinners based on the ads we show them. In addition to click prediction, we look at how likely a user is to save or hide an ad. We make these predictions for different types of ad formats (image, video, carousel) and in context of the user (e.g., browsing the home feed, performing a search, or looking at a specific Pin.)

咖啡廳裡親子的對話

 有一天，在咖啡廳裡工作，聽到隔壁一對父母在教小孩。

大概的情況是，小孩子的作業訂正有問題，媽媽非常地生氣，期間也說明別人是如何如何，為什麼你都做不好？小孩子低聲啜泣，對媽媽的話，也不太敢反駁。爸爸則是說，我們不是和乞丐比較，唸書是要幫你找到好工作、賺錢 。後來，我和小孩剛好都離座，在走道上，我仔細地看了小孩一眼，一看就是一個乖巧、懂事的小孩子。回座後，夫妻開始有不同的意見。前後超過一個小時。

當下有非常高的衝動，想要和父母講一下。但是，我怕被白眼，所以還是忍了下來。