William Strunk Jr. and E.B. White, The Elements of Style, Pearson, 1999.
英文寫作聖經，值得一讀再讀。
4/30/2019
4/27/2019
5 of the biggest threats to our oceans
United Nations Development Programme (UNDP), 2017/6/30 (5 videos)
4/20/2019
Artificial Intelligence: A Modern Approach
Stuart Russell and Peter Norvig, Artificial Intelligence: A Modern Approach, 3/e, Prentice Hall, 2009.
完整且經典的書。Python 碼。
On a Formal Model of Safe and Scalable Self-driving Cars (自駕車)
Shai Shalev-Shwartz, Shaked Shammah, and Amnon Shashua, On a Formal Model of Safe and Scalable Self-driving Cars, arXiv:1708.06374, Mobileye, 2017.
In order to gain perspective over the typical values for such probabilities, consider public accident statistics in the United States. The probability of a fatal accident for a human driver in 1 hour of driving is 10^−6. From the Lemma above, if we want to claim that an AV meets the same probability of a fatal accident, one would need more than 10^6 hours of driving. Assuming that the average speed in 1 hour of driving is 30 miles per hour, the AV would need to drive 30 million miles to have enough statistical evidence that the AV under test meets the same probability of a fatal accident in 1 hour of driving as a human driver.... (*)
4/19/2019
Risk-based policies for airport security checkpoint screening (機場安檢檢查站檢查)
L.A. McLay, A.J. Lee, and S.H. Jacobson, Risk-based policies for airport security checkpoint screening, Transportation Science, Volume 44, Issue 3, August 2010, pp. 333-349. (Informs 2018 Impact Prize)
Passenger screening is an important component of aviation security that incorporates real-time passenger screening strategies designed to maximize effectiveness in identifying potential terrorist attacks. This paper identifies a methodology that can be used to sequentially and optimally assign passengers to aviation security resources. An automated prescreening system determines passengers' perceived risk levels, which become known as passengers check in. The levels are available for determining security class assignments sequentially as passengers enter security screening. A passenger is then assigned to one of several available security classes, each of which corresponds to a particular set of screening devices. The objective is to use the passengers' perceived risk levels to determine the optimal policy for passenger screening assignments that maximize the expected total security, subject to capacity and assignment constraints. The sequential passenger assignment problem is formulated as a Markov decision process, and an optimal policy is found using dynamic programming. The general result from the sequential stochastic assignment problem is adapted to provide a heuristic for assigning passengers to security classes in real time. A condition is provided under which this heuristic yields the optimal policy. The model is illustrated with an example that incorporates data extracted from the Official Airline Guide.
Temporal Big Data for Tactical Sales Forecasting in the Tire Industry
Yves R. Sagaert, El-Houssaine Aghezzaf, Nikolaos Kourentzes, and Bram Desmet, Temporal Big Data for Tactical Sales Forecasting in the Tire Industry, Interfaces, Volume 48, Issue 2, March-April 2018, pp. 121–129.
We propose a forecasting method to improve the accuracy of tactical sales predictions for a major supplier to the tire industry. This level of forecasting, which serves as direct input to the demand-planning process and steers the global supply chain, is typically done up to a year in advance. The product portfolio of the company for which we did our research is sensitive to external events. Univariate statistical methods, which are commonly used in practice, cannot be used to anticipate and forecast changes in the market; and forecasts by human experts are known to be biased and inconsistent. The method we propose allows us to automate the identification of key leading indicators, which drive sales, from a massive set of macroeconomic indicators, across different regions and markets; thus, we can generate accurate forecasts. Our method also allows us to handle the additional complexity that results from short-term and long-term dynamics of product sales and external indicators. For the company we study, accuracy improved by 16.1 percent over its current practice. Furthermore, our method makes the market dynamics transparent to company managers, thus allowing them to better understand the events and economic variables that affect the sales of their products.
Improving U.S. Navy Campaign Analyses (海軍戰役分析) with Big Data
Brian L. Morgan, Harrison C. Schramm, Jerry R. Smith, Jr., Thomas W. Lucas, Mary L. McDonald, Paul J. Sanchez, Susan M. Sanchez, Stephen C. Upton, Improving U.S. Navy Campaign Analyses with Big Data, Interfaces, Volume 48, Issue 2, March-April 2018, pp. 130–146.
Decisions and investments made today determine the assets and capabilities of the U.S. Navy for decades to come. The nation has many options about how best to equip, organize, supply, maintain, train, and employ our naval forces. These decisions involve large sums of money and impact our national security. Navy leadership uses simulation-based campaign analysis to measure risk for these investment options. Campaign simulations, such as the Synthetic Theater Operations Research Model (STORM), are complex models that generate enormous amounts of data. Finding causal threads and consistent trends within campaign analysis is inherently a big data problem. We outline the business and technical approach used to quantify the various investment risks for senior decision makers. Specifically, we present the managerial approach and controls used to generate studies that withstand scrutiny and maintain a strict study timeline. We then describe STORMMiner, a suite of automated postprocessing tools developed to support campaign analysis, and provide illustrative results from a notional STORM training scenario. This new approach has yielded tangible benefits. It substantially reduces the time and cost of campaign analysis studies, reveals insights that were previously difficult for analysts to detect, and improves the testing and vetting of the study. Consequently, the resulting risk assessment and recommendations are more useful to leadership. The managerial approach has also improved cooperation and coordination between the Navy and its analytic partners.
The New York City Off-Hour Delivery Program
Jose Holguin-Veras, et al., The New York City Off-Hour Delivery Program: A Business and Community-Friendly Sustainability Program, Interfaces, Volume 48, Issue 1, January-February 2018, pp. 70–86.
The New York City Off-Hour Delivery (NYC OHD) program is the work of a private-public-academic partnership—a collaborative effort of leading private-sector groups and companies, public-sector agencies led by the New York City Department of Transportation, and research partners led by Rensselaer Polytechnic Institute. The efforts of this partnership have induced more than 400 commercial establishments in NYC to accept OHD without supervision. The economic benefits are considerable: the carriers have reduced operational costs and parking fines by 45 percent; the receivers enjoy more reliable deliveries, enabling them to reduce inventory levels; the truck drivers have less stress, shorter work hours, and easier deliveries and parking; the delivery trucks produce 55–67 percent less emissions than they would during regular-hour deliveries, for a net reduction of 2.5 million tons of CO2 per year; and citizens’ quality of life increases as a result of reduced conflicts between delivery trucks, cars, bicycles, and pedestrians, and through the use of low-noise delivery practices and technologies that minimize the impacts of noise. The total economic benefits exceed $20 million per year. The success of the OHD program is due largely to the policy design at its core, made possible with the behavioral microsimulation. This unique optimization-simulation system incorporates the research conducted into an operations research/management science tool that assesses the effectiveness of alternative policy designs. This enabled the successful implementation of the project within the most complex urban environment in the United States.
Barco Implements Platform-Based Product Development in Its Healthcare Division
Robert N. Boute, Maud M. Van den Broeke, and Kristof A. Deneire, Barco Implements Platform-Based Product Development in Its Healthcare Division, Interfaces, Volume 48, Issue 1, January-February 2018, pp. 35–44.
In this article, we present how Barco, a global technology company, used an operations research optimization model, which was supported by an efficient solution method, to implement platforms—common structures from which sets of products could be made—for the design and production of its high-tech medical displays. Our optimization model captures all cost aspects related to the use of platforms; thus, it is an objective tool that considers the input from marketing, sales, research and development (R&D), operations, and the supply chain. This comprehensive view allowed Barco to avoid the excessive costs that may result from the implementation of an incorrect platform. Our model supported Barco in determining the elements that should comprise each platform, the number of platforms to develop, and the products to derive from each platform. The results of the project led to reductions in safety stock and increased flexibility due to the use of platforms: R&D can now introduce twice as many products using the same resources, thus increasing Barco’s earnings by more than five million euros annually and reducing product introduction time by nearly 50 percent.
Discrete-Event Simulation Modeling Unlocks Value for the Jansen Potash Project
Sylvie C. Bouffard, Peter Boggis, Bryan Monk, Marianela Pereira, Keith Quan, Sandra Fleming, Discrete-Event Simulation Modeling Unlocks Value for the Jansen Potash Project, Interfaces, Volume 48, Issue 1, January-February 2018, pp. 45–56.
BHP plans to enter the bulk fertilizer market by developing its first potash operation, the Jansen Potash Mine, in Saskatchewan, Canada. In conjunction with Amec Foster Wheeler, the Jansen project team developed a model of the Jansen production and logistics chain to understand the drivers of production capacity. The Detailed Integrated Capacity Estimate model (DICE) is a comprehensive discrete-event simulation model of Jansen’s upstream production (mining, hoisting, and ore processing) and downstream logistics (rail, port, and marketing). DICE provides an unprecedented combination of complexity, granularity, and scalability, which informs ore storage capacities, product sizing infrastructure, critical-equipment redundancies, bypasses, and operational practices. The team used DICE during the prefeasibility study of the Jansen project. The model provided the justification for the removal of about $300 million in capital expenses to equip the second of two hoisting shafts, the reduction of planned maintenance, and the increase of the degree of mining automation. Throughout the prefeasibility study, Jansen’s annual production in stage 1 of operations was estimated to increase by 15–20 percent, with two-thirds of this gain credited to DICE. This potential additional production added $500 million to the net present value of Jansen stage 1. In consideration of this, among other factors, the BHP board of directors approved the transition of the Jansen project from a prefeasibility to a feasibility study.
A Novel Movement Planner System for Dispatching Trains
Srinivas Bollapragada, Randall Markley, Heath Morgan, Erdem Telatar, Scott Wills, Mason Samuels, Jerod Bieringer, Marc Garbiras, Giampaolo Orrigo, Fred Ehlers, Charlie Turnipseed, Jay Brantley, A Novel Movement Planner System for Dispatching Trains, Interfaces, Volume 48, Issue 1, January-February 2018, pp. 57–69.
General Electric Company (GE) partnered with Norfolk Southern Railroad (NS) to create and implement an optimization algorithm-based software system that dispatches thousands of trains in real time, increases their average speed, and allows NS to realize annual savings in the hundreds of millions of dollars. NS handles a range of rail traffic that includes intermodal, automobile transport, manifest freight, and passenger, all with unique priorities and scheduling requirements. Previously, dispatching for each geographic area was managed manually from regional dispatch centers and did not encompass a view of the entire rail network. The algorithm that we developed incorporates data about the properties of the rail networks (e.g., track layout, speed restrictions, height and weight restrictions), data about the trains (e.g., schedules, operating costs, train characteristics), and additional activities associated with train dispatching, such as crew changes and inspections. In doing so, we created a novel system to manage all train dispatching, increased the average speed of trains by two miles per hour, and decreased operating costs, while significantly improving schedule adherence and crew expirations. Every mile-per hour increase in average speed translates to $200 million savings in capital and operational expenses annually for NS. GE is currently implementing this system at two other railroads and is gaining additional important benefits from the project.
American Red Cross Uses Analytics-Based Methods to Improve Blood-Collection Operations
Turgay Ayer, Can Zhang, Chenxi Zeng, Chelsea C. White III, V. Roshan Joseph, Mary Deck, Kevin Lee, Diana Moroney, Zeynep Ozkaynak, American Red Cross Uses Analytics-Based Methods to Improve Blood-Collection Operations, Interfaces, Volume 48, Issue 1, January-February 2018, pp. 24–34.
In this study, we describe a regional-level cryoprecipitate (cryo)-collection project at the American Red Cross Southern Region, one of the 36 Red Cross regions in the United States, which serves more than 120 hospitals in the Southern part of the country. Managing collections for cryo units is particularly challenging because producing cryo requires the collected whole blood to be processed within 8 hours after collection; for all other blood products, this time constraint is at least 24 hours. This project focuses on dynamically determining when and from which mobile collection sites the American Red Cross Southern Region should collect whole blood for cryo production, such that it meets its weekly collection targets and minimizes its collection costs. To solve this problem, we developed a new collection model, which allows different types of collections at the same collection site and developed a dynamic programming approach to solve the problem to near optimality. Analyzing the dynamic programming results led us to create a greedy-algorithm heuristic, which we implemented in a decision support tool (DST) to systematize the selection of the collection sites. The implementation of the DST in the Red Cross Southern Region resulted in an increase in the number of whole blood units that can be shipped back to the production facility and processed within eight hours after collection. During the fourth quarter of 2016, this facility processed about 1,000 more units of cryo per month (an increase of 20 percent) at a slightly lower collection cost, resulting in an approximately 40 percent reduction in the per-unit collection cost for cryo. Based on the successful implementation in the Southern Region, the American Red Cross also implemented our DST in its St. Louis facility and plans to implement it at its 10 other cryo production facilities.
4/14/2019
Energy Companies Using AI for Cost-Efficiency
Ellen Chang, 5 Energy Companies Using AI for Cost-Efficiency, US News, April 12, 2019.
Companies use AI to detect faults such as cracks in pipelines and machinery by analyzing images. This saves companies money and minimizes downtime when equipment breaks down, says Guido Jouret, chief digital officer of ABB, a Swiss power and automation company.
"An AI pilot project with one of the world's largest hydroelectric utilities showed a 10% reduction in routine maintenance and a 2% increase in output," he says. "These measures translate into millions of dollars in cost savings."...
Schneider Electric leverages Microsoft Corp.'s (MSFT) machine learning capabilities to monitor and configure pumps in the oil and gas field remotely since early detection of a pump failure can avoid weeks of the equipment being out of commission and repair costs of up to $1 million.
"Our customer Tata Power, India's largest power generator saved almost $300,000 on a single predictive maintenance catch," he says.
Gogoro 2019 大佈局
Chris，新獲 72 億聯貸案，Gogoro 跟我們談了 2019 大佈局，Inside，2019/04/03
「台灣的電不乾淨嗎？」陸學森首先解釋一些關於電力的迷思。
事實上台灣只有 4% 的 PM2.5 來自電廠發電，然而有 25% 的 PM 2.5 來自汽機車排放；電動機車能源效率比較差嗎？錯了，燃油機車一公升最多只能跑 38 公里，但用在發電再儲存到電動機車，足足可以跑 85 公里。
「那電動機車會不會很耗電嗎？」但實際上一天的家庭總用電，其實就足夠給電動機車騎22天。電動機車平均一天騎起來只用0.56度，算起來就算全台1386萬台機車都變電動車了，也只佔台灣一天發電量 1.2%。但更特別的是，Gogoro 電網也跳開了全台一般的用電習慣，在晚上才充電去平衡電網需求。...
「初期有一大部分馬達就是士林電機幫我們做的。」陸學森也很自豪地說，Gogoro 高技術的馬達與零件也對台灣生產鏈起了升級作用。目前 Gogoro 自有生產線就有 48 組機器手臂、30 道工作模組，確實落實了工業 4.0 的概念。
談到換電，目前他們在台灣投資超過 105 億用於智慧電網裡，現在台灣 1200 站，預計年底前要鋪到 1500 站，完成全台車主平均 2 公里內就有站可以換電的目標。
4/12/2019
Ethics guidelines for trustworthy AI by the European Commission
Ethics guidelines for trustworthy AI, REPORT / STUDY8 April 2019
According to the guidelines, trustworthy AI should be:
(1) lawful - respecting all applicable laws and regulations
(2) ethical - respecting ethical principles and values
(3) robust - both from a technical perspective while taking into account its social environment
The guidelines put forward a set of 7 key requirements that AI systems should meet in order to be deemed trustworthy. A specific assessment list aims to help verify the application of each of the key requirements.
AI-Based Massively Multivariate Conversion Rate Optimization (轉換率最佳化)
R. Miikkulainen, et al., Sentient Ascend: AI-Based Massively Multivariate Conversion Rate Optimization, 2018 AAAI.
Conversion rate optimization (CRO) means designing an ecommerce web interface so that as many users as possible take a desired action such as registering for an account, requesting a contact, or making a purchase. Such design is usually done by hand, evaluating one change at a time through A/B testing, or evaluating all combinations of two or three variables through multivariate testing. Traditional CRO is thus limited to a small fraction of the design space only. This paper describes Sentient Ascend, an automatic CRO system that uses evolutionary search to discover effective web interfaces given a human-designed search space. Design candidates are evaluated in parallel on line with real users, making it possible to discover and utilize interactions between the design elements that are difficult to identify otherwise. A commercial product since September 2016, Ascend has been applied to numerous web interfaces across industries and search space sizes, with up to four-fold improvements over human design. Ascend can therefore be seen as massively multivariate CRO made possible by AI.