San Francisco 2014
 
Blog:
blog home Home
 
       Follow:
RSS
 
Nov 11 14

INFORMS Connect: Facebook for ORMS

by Tapas Das

INFORMS Connect: The Information Exchange and Marketplace

Tapas K. Das

Yes, Facebook competitor is here, the INFORMS Connect. You can find your friends and colleagues, see their profile and activities, and share news and information via emails to specific groups. Soon Connect will be the job placement information sharing hub where all the seekers and employers will converge. Connect will also offer the capability to have group meetings, among many others.

Like any other marketplace (remember healthcare.gov? lol), INFORMS Connect is not perfect yet.  Some in the test group have experienced a large volume of emails from Connect. Mary Leszczynski, the Connect coordinator made a presentation at the meeting of the Council of Industrial Engineering Academic Department Heads (CIEADH) on Sunday (see picture). She explained means of adjusting the preferences to manage the email flow. Mary is seeking our feedback on the functionalities. Please visit the connect website http://connect.informs.org/home and contact Mary with your feedback at mary.leszczynski@informs.org.

Mary Leszczynski_INFORMS Connect

Nov 10 14

INFORMS early morning idea for Tuesday

by Warren Lieberman

Here’s a thought for everyone who is looking for their early morning coffee prior to the 8 am technical session on Tuesday. Rather than wait on a line of 40 people or so, if you leave the hotel, turn right, and walk about a block or two, there are two Starbucks locations. The one across the street had a line of 4 people at 7:15 with plenty of food choices as well.

There are plenty of options. Don’t be afraid to explore! And perhaps going somewhere different will put you in in a better frame of mind as you think of all the different opportunities that the technical sessions can lead to

Nov 10 14

JFIG Paper Competition Results

by Michael Trick

The Junior Faculty Interest Group paper competition is one of the highlights of the INFORMS conference for me.  I was part of the (large) judging panel for this year, and there were many, many great papers submitted.  It made it very difficult to get down to a final set, let alone pick a winner.  But the panel did end up both with six finalists and with a selection of first, second and third place winners.  Here are the results:

 

First Place:
Two-Term Disjunctions on the Second-Order Cone
Fatma Kilinc-Karzan and Sercan Yildiz

Second Place:
Unbalanced Random Matching Markets: The Stark Effect of Competition
Itai Ashlagi, Yashodhan Kanoria, and Jacob Leshno

Third Place:
Third Place: Multistage Adaptive Robust Optimization for Electric Power Systems Operation
Andy Sun and Alvaro Lorca

Honorable Mention:
Optimal Learning with Non-Gaussian Rewards
Zi Ding and Ilya Ryzhov

Modified Echelon (r, Q) Policies with Guaranteed Performance Bounds for Stochastic Serial Inventory Systems
Ming Hu and Yi Yang

A Large-Scale Multi-Server Fork-Join Network with Non-Exchangeable Synchronization
Hongyuan Lu and Guodong Pang

Congratulations to all of the participants!

 

Nov 10 14

Alvin Roth’s Plenary Talk

by Chun Ye

It is 9:45am, I am navigating my way through the meeting rooms in Hilton, like a mouse in a maze. My destination: Alvin Roth’s plenary talk. As a student who is interested in market design, I have been looking forward to hear the words of wisdom of this Nobel prize winning economist at this year’s INFORMS Annual Meeting. I remember searching for the abstract of Al Roth’s talk inside the giant directory given to me at registration. To my dismay, his talk has no abstract other than the intriguing title “Market Design & the Economist as Engineer.” The lack of an abstract only adds more suspense to his impending talk, or so I thought.

I arrive at the Continental room 10 minutes prior to the start of the session with the intention of claiming a prime viewing location in the audience, only to find out that the majority of them are claimed by those who have arrived earlier. As a student trained in operations research who is used to taking a shortest path between a source destination pair, I realize that my back of the envelop calculations failed to take into consideration the human factors involved: everyone else chose to take the shortest path as well. It turns out that I (as well as others) experienced firsthand the kind of optimization problems that Al Roth is about to address in his talk.

Al Roth starts off his talk by justifying why an economist is invited to speak in front of an audience of operations researchers: “economist is a label that people associate me with, but I am an engineer at heart.” Al Roth received his PhD from what used to be the OR department in Stanford in 1974. During his PhD days, most people in the department worked on problems involving optimizing the operations of a firm. Many such problems involved a single decision maker. In contrast, Al Roth was interested game theoretic problems, in which multiple decision makers are involved. Game theory was a subject of studies by economists, who were primarily interested in understanding the long run equilibrium behavior of an existing system. Al Roth, on the other hand, was more interested in designing a system that will influence the decisions of each individual agent in the system towards a behavior that is desired by the central planner. In that sense, the term “the economist as engineer” used in Al Roth’s talk title is not an oxymoron. In fact, Al Roth dedicated his career implementing and preaching the philosophy of using engineering techniques to design exchange markets. In doing so, he revolutionized the field of market design.

In his talk, Al Roth presents three applications of market design: the medical labor market, school choice, and kidney exchange. Just as congestion was created by everyone taking the shortest path to attend Al Roth’s plenary presentation, which results some degree of inefficiency, a natural congestion forms in each of the aforementioned applications, whether it be doctors wanting to get into the most popular hospital, students the best schools, or patients a functional kidney. Al Roth explains that under certain simplifying assumptions, beautiful theories developed in the works of Gale and Shapley, Edmond and Gallai, and others can be turned into mechanism that a central planner uses to improve the efficiency of these market exchanges. Unfortunately, beautiful theories do not always work as intended in real-world settings as some of the underlying assumptions that give backbone to these theories are usually not satisfied. Consequently, it is the role of an engineer to monitor how well these mechanisms work in practice, to identify the deficiency of such mechanisms, and to tweak the mechanism in small ways so as to make up for its shortcomings.

For instance, the assignment of doctors to hospitals can be thought of as a two-sided matching problem. In this problem, each hospital wants to hire a number of doctors and each doctor wants to get a job at a hospital. We assume that agents on each side of the market has a preference ordering over members of the opposite site. The Deferred Acceptance algorithm (DA) proposed by Gale and Shapley in the 1960s is a standard method for finding a matching that every agent in the market find acceptable. The standard notion of acceptability used in the two sided matching literature is the stability condition, which states that no pairs of hospital and doctor that are not matched to each other will be better off by choosing to abandon their current partner and get together. The DA algorithm works by having each doctor iteratively propose to their top ranked hospital that has not (yet) rejected him. Each hospital would temporarily hold on to the proposals that it receives until its capacity is fill, in which case it will start rejecting a previously accepted candidate that it finds the least attractive provided that it receives a proposal from a more attractive candidate. The algorithm terminates when either all doctors are matched or all of the unmatched doctors have proposed to every hospital that they find acceptable. In the context of two sided matching, the DA algorithm is guaranteed to return a stable matching and as a result, it quickly became one of the most popular methods for assigning doctors to hospitals in many places around the world.

Nonetheless, in the 1970s, people who run the allocation mechanism started to notice that some people did not show up at the hospital that was assigned to them by the DA algorithm but rather arranged matching with another hospital in a decentralized fashion. In fact, most of the doctor-hospital blocking pairs of involved a doctor who was married to another doctor and the couple jointly participated in the allocation mechanism. Al Roth sums up the “couples’ dilemma” nicely: “the iron law of marriage: You can’t be happier than your spouse.” The existing DA algorithm suffers from the drawback of its assumption: every doctor’s preference over hospitals is independent of the assignment outcome of every other doctor. In the case of couples, it makes more sense for the couple to jointly report a preference ordering over a pair of hospitals, rather than two separate preference orderings over individual hospitals, as their individual assignment does affect the preference of their partner. In the variant of the “couples” two-sided matching problem, a stable matching may not exist. However, it has been observed by practitioners empirically that a stable matching in many large matching markets do exist in real life, and a stable matching can still be computed by a variant of the DA algorithm efficiently. In a recent joint work, Al Roth and several other researchers were able to prove formally, under certain technical assumptions, that a stable matching in large (random) markets exists with high probability. Hence, their theoretical result support the existing empirical findings.

Another application that Al Roth has been actively involved in is the field of kidney exchange. Kidney exchange is a 70 billion industry in the U.S. with 100,000 patients currently on the waiting list for a healthy kidney. Since a family member of a patient cannot always directly donate a kidney to the patient due to incompatibility of blood types and other complications, the scarcity of the supply of kidneys motivates the design of an efficient market for patients and their family members to increase the supply of kidneys that otherwise would have to come from deceased donors. Due to ethical reasons, many countries forbid monetary transactions for kidney, this leads to the natural idea of having a patient-donor pairs exchange kidneys with each other. Unfortunately, even the idea of kidney exchange has its own complications. For instance, the size of the kidney trading cycles involved in an exchange of patient-donor pairs must be small (as in 2 or 3), else hospital will not have enough resources to simultaneously perform the necessary medical operations. Maximizing the number of kidney exchanges involving small trading cycles is a NP-hard problem in general. However, Al Roth and colleague provided some empirical findings and theoretical models to explain why some of algorithms developed work better in practice than the theoretical bounds that one is able to derive. Recently, Al Roth and colleague have been quantifying the effect of exchanges initiated from a voluntary donor. Empirical findings as well as theoretical models are developed to address and explain such an effect.

Al Roth concludes the talk by encouraging operations researchers to collaborate with economists and computer scientists in addressing some of difficult problems that arise in designing an efficient protocol for a multi-agent system. Operation researchers are adapt at designing and optimizing a system involving a single decision maker. However, many real world settings involve multiple agents who simultaneously making distinct decisions, and being able to take these decisions into consideration is the key in designing a functional, robust, and long lasting system. Al Roth has done some great work throughout his career: he has kept marriages stable, put smiles on parents’ faces, and saved countless lives. Al Roth makes the world a better place.

Nov 10 14

Navigating INFORMS 2014

by Paul Rubin

Sung to the chorus of “I Saw Linda Yesterday“:

I went up, down,
Like a merry-go-round and round,
Like a falling star down, down,
Oh yeah,
I got lost while on my way.

Nov 10 14

INFORMS: Perspectives – Peter Mayoros and Elizabeth Olin on Healthcare Scheduling

by David Morrison

The first time I attended INFORMS was in Washington D.C. in 2008; perhaps the unusual part here is that I came as an undergrad, presenting some work that I did with my advisor, Susan Martonosi (spelled with an ‘o’). It was definitely an intimidating experience to be at this enormous conference, and everyone was talking about things I didn’t understand — but it was also exhilirating! And, though I didn’t realize it then, it would prove very beneficial for my long-term research career: right now I’m pursuing some collaborations that have grown out of my undergraduate research, all because I was able to come to INFORMS and present my work. So today’s INFORMS: Perspectives comes from two students who are just like I was 6 years ago.


peter_elizabeth_small

Elizabeth Olin and Peter Mayoros are two senior undergradute students at the University of Michigan, Ann Arbor, working with Amy Cohn in the Center for Healthcare Engineering and Patient Safety (CHEPS). Actually, it turns out that Elizabeth and Peter are part of a contingent of 12 students from CHEPS (and a host of others from UMich), and the research output from this group is prolific. There’s an astonishing number of talks being given on various and sundry healthcare topics at INFORMS this year, and many of them from CHEPS!

Elizabeth and Peter are both working on topics in healthcare scheduling, which is a very important, but difficult area. Elizabeth is developing schedules for surgeons at the University of Colorado Orthopedics department, whereas Peter is dealing with scheduling medical residents in pediatric residency programs. They’re using techinques that will be familiar to many people here — modeling the problem as an integer program, and then using standard techniques to produce a feasible schedule. In fact, both Elizabeth and Peter commented on how encouraging it is to come to INFORMS and see other people applying similar techniques to other problems. Elizabeth says, “There are so many new ideas to soak in here!” She’s trying to figure out “what she can gain from every one of these talks” that can be applied to her current research. Peter had similar things to say: it’s extremely helpful to see what other people are doing, “particularly if you’re at a dead end.”

And really, that’s the point of INFORMS. To meet other people and learn from them. To get such an opportunity as an undergraduate is, as many can attest, eye-opening. Both Elizabeth and Peter extolled the CHEPS program at UMich for providing them with such good research experience. “Not every undergraduate gets to do research,” says Elizabeth, and at CHEPS she’s had the opportunity to work not only with other undergraduates, but also master’s students, PhD students, postdocs, facutly, premed students, doctors, nurses, hospital staff—the list goes on and on. She’s had more real observation time in hospitals than many students would ever dream of.

The work doesn’t come without it’s challenges. Doing research as an undergraduate, when you only have 8-12 hours a week to work on a problem, is pretty tough. Peter commented that it’s difficult to even know where to start on some of these problems, and the sheer size of the problem is intimidating. Fortunately, there’s lots of other people around to support Elizabeth and Peter in their research (are you sensing a common theme here yet?). Having other people around is also helpful for navigating the conference; they spoke of looking to “experienced INFORMers” to provide guidance on where to go next and what to do at the conference.

I asked what their plans were post-graduation, and they both have some pretty exciting prospects. For Elizabeth, she’s keeping her options open, but she loves what she’s doing and graduate school is definitely in the cards. Next year, Peter is planning to work with Youthworks-Detroit, a non-profit organization that provides mentorship opportunities for inner-city youth in Detroit (awesome!). After that, who knows? But it will definitely involve healthcare engineering, he says.

In any case, they’ve been having a blast in San Francisco and at INFORMS this year, and there’s still a lot to do! Elizabeth is giving a talk entitled Predicting Disposition for Pediatric Asthma Patients in the Wednesday morning session (WA40), and Peter is speaking about his work in Block Scheduling for a Pediatric Residency Program today at 1:30 (MC38) — so if you want to know more, make sure to attend their talks, or flag them down in the hallways! Best of luck, and we’ll expect to see you both back as “experienced INFORMers” next year!

Nov 10 14

Edelman reprise

by Harrison Schramm

Polio eradication. Now in continental 6 ballroom.

Nov 10 14

Intel’s portfolio management

by Patrick Leach

I just attended Intel’s presentation on their approach to portfolio management.  Their portfolio is more complex than most to manage because 1) so many of their projects are interrelated (you can’t do project A without first doing project B, project C enhances project D, etc.), 2) projects involve the commitment of significant resources, so once you start one, it’s not trivial to stop, and 3) many of their products are only on the market for a couple of years (or less) before being replaced by something better.

As a result, they have adopted what I think of as the Deep Blue approach to portfolio management.  Deep Blue was IBM’s chess playing computer (or one of them – there was a series of them).  It’s approach to chess was to start with every possible combination of moves, and then whittle it down to the one that made it most likely to achieve its objectives (e.g., controlling space on the board or capturing a key piece in the early part of the match, achieving checkmate in the latter part of the match).  When Deep Blue played Garry Kasparov in a rematch in 1997, Deep Blue won.

Likewise, Intel has created an algorithm that starts by generating every possible portfolio, cuts out the impossible ones, reduces the set down to non-dominated ones (the efficient frontier), and then applies things like budgetary constraints and strategic direction.  This is a change from most portfolio management, which often starts with strategic direction and then generates a manageable number of portfolios to evaluate and learn from, with the idea that you use what you learn from the early analyses to create a portfolio that is better than anything you came up with originally.

I was pleased to see that Intel’s approach does not just punt the portfolio decision to the calculations, though.  As they put it, “The analysis informs the intuition, and the intuition informs the analysis.”  It’s an iterative process between assessing, modifying, and reassessing various possible portfolios based on what senior management wants and believes, and also putting senior management in the position of reassessing their instincts and strategic direction based on the results of the analyses.  So it’s sort of like letting Garry Kasparov use Deep Blue to inform his decisions during the chess match.

It was a pretty cool presentation.

Pat Leach

Nov 10 14

Serendipity and INFORMS San Francisco

by Anna Nagurney

I love surprises and especially when they are wonderful!

One of the great aspects of this INFORMS conference Is the people that you meet just due to pure serendipity. With the number of registrants over 5,200 one can sometimes feel overwhelmed but it is those special unplanned meetings when one sees colleagues and even students from long ago that one remembers and treasures.

Yesterday, while waiting for a student for the Coffee with a Member initiative in the Hilton lobby, I ran into a colleague from Europe that I had not seen in years. He was looking for a former colleague from MIT and, although our scheduled meetings never took place, we had a great conversation!

This morning, who was in the elevator but a former student from UMass Amherst who is now at VT! And while wandering the streets around the hotel – yes, we east coast folks are up very early – I saw one of my former PhD students who works at SAS and I had no idea that she was even here. The exclamations of joy and hugs that followed elicited lots of stares from onlookers. Those wearing their INFORMS badges were just smiling with recognition.

INFORMS is now a big professional society with over 11,000 members but it is the special communities that make it feel personal

Enjoy the chance meetings and the making of many fabulous memories!

Nov 10 14

50th Anniversary of Decision Analysis

by Freeman Marvin

Howard Raiffa was the guest of honor at Saturday’s DA gala celebration.

2014-11-08 18.22.37