Schedule for: 17w2670 - Future Research Directions in Digital Simulation Methodology for the Next 10 Years

Note: the Lecture rooms are available after 16:00.

For meal options at the Banff Centre, there are food outlets on the campus such as Vistas Main Dining Room on the 4th floor of Sally Borden Building (breakfast: 7:00-9:30am; lunch: 11:30am-1:30pm; dinner: 5:30-7:30pm), Le Cafe (ground floor, Sally Borden Building) and the Maclab Bistro (Kinnear Centre). You will also find a good selection of restaurants in the town of Banff which is a 10-15 minute walk from Corbett Hall.

Beverages and a small assortment of snacks are available in the lounge on a cash honour system.

A buffet breakfast is served daily between 7:00am and 9:00am in the Vistas Dining Room, the top floor of the Sally Borden Building. Note that BIRS does not pay for meals for 2-day workshops.

On what should we work so that simulation remains relevant to users, vendors, researchers and consumers in the future? This is a harder question than what we should do so that simulation stays interesting to us. To this captive focus group I will present my musings on three analysis challenges that I assert would also be relevant to everyone else, as possible topics for my WSC ’17 keynote address. • All models are wrong, but some models are wronger. • Simulation: the glitter or the glue? • All the world's a database; and all the models merely data sources.

In March 2016 in Seoul, Korea, Google DeepMind’s AlphaGo, a computer Go-playing program, defeated the reigning human world champion Go player, a feat far more impressive than previous computer programs victories in chess (Deep Blue) and Jeopardy (Watson). Due to the sheer combinatorial nature of the number of possibly game configurations, at the heart of all computer Go-playing algorithms is Monte Carlo tree search, which is based on an upper confidence bound (UCB) algorithm that traces its roots back to an adaptive multi-stage sampling algorithm for estimating the value function in finite-horizon Markov decision processes (MDPs), introduce in a paper published in Operations Research in 2005. We describe this algorithm, the main ideas behind AlphaGo, and speculate on potential ways the OR and AI communities could benefit from more research cross-fertilization on the stochastic simulation side.

A buffet lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building. Note that BIRS does not pay for meals for 2-day workshops.

Meet in foyer of TCPL to participate in the BIRS group photo. The photograph will be taken outdoors, so dress appropriately for the weather. Please don't be late, or you might not be in the official group photo!

Take a short walk to Lake Louise or Banff downtown and have some fun.

Many modern "machine learning" and "big data" optimization problems can be seen as variations of the traditional simulation optimization problem which is well-known within the simulation community. While stochastic gradient descent (known to many of us as "stochastic approximation") has been the dominant solution method within machine learning and big data, there is a nascent but unmistakable trend of using more sophisticated step-length and direction finding mechanisms, e.g., line search and trust region, along with appropriate sampling strategies. In this talk, towards solving optimization problems having a smooth objective function that can be observed with an inexact oracle, I will outline a method that combines line search and adaptive sampling. This general philosophy of using a line search with adaptive sampling, appropriately modified to account for constraints and the nature of the feasible region, has proven expedient in a wide variety of contexts we have encountered recently. I will present "Nesterov type" complexity results for the proposed method for the context where the inexact oracle is quasi-Monte Carlo or numerical integration.

I’ll discuss a number of issues surrounding ranking and selection algorithms and their use in simulation. I’ll argue that we should prefer “probably approximately correct” (PAC) selection (also known as “good” selection) over the dominant “correct” selection, and explore when correct selection actually implies PAC selection. I’ll then discuss the use of selection methods to clean up after search. I’ll provide a new class of PAC selection methods that appear to be more efficient than existing ones, and finally discuss the need for, and methods of, predicting running times of these procedures as we scale to large parallel computing platforms.

Please join us. Payment for dinner will be the responsibility of the participants. However, a cake and sparkling wine will be provided on the behalf of the IEMS faculty in Northwestern University and some wine will be provided courtesy of Barry's previous students.

2-day workshop participants are welcome to use BIRS facilities (Corbett Hall Lounge, TCPL, Reading Room) until 15:00 on Sunday, although participants are still required to checkout of the guest rooms by 12 noon. There is no coffee break service on Sunday afternoon, but self-serve coffee and tea are always available in the 2nd floor lounge, Corbett Hall.

Raghu Pasupathy will lead a discussion on three items which are of interest to the simulation society.

In this talk, we will discuss research questions that are likely to become more important in the next decade, as data, modeling, and simulation become more tightly integrated within various decision platforms. Simulation can play a key role in improving decision-making, but the ways in which simulation will be used may require new research-based methodologies.

Russell Barton, John Fowler, Susan Sanchez, Bruce Schmeiser, Lee Schruben