Thanks for visiting! I have moved my blog. An updated version of this post can be found here.
(As always, this post reflects my own views and not necessarily those of Microsoft.)
I recently returned from the INFORMS conference on operations research in Austin, Texas. INFORMS is the most popular professional organization relating to the theory and practice of operations research - which includes numerical optimization, metaheuristic search techniques, and (increasingly) analytics. Their annual conference is on the calendar of nearly all the top researchers and software vendors in the field. My first INFORMS conference was nearly 10 years ago and I can definitely say the conference is much larger than it was back then. There were 75 technical tracks, two full days of software demonstrations, as well as special interest groups and workshops. Attendees generally fall into four categories: students, researchers, practitioners, and software vendors. I represented the Microsoft Technical Computing team at the conference, but several of my colleagues in other parts of the company, including Microsoft Research, also attended. Microsoft's presence was stronger than in past years. We sponsored the INFORMS awards dinner, my colleague Gunther participated in a roundtable on cloud computing, I manned a booth on the conference floor, and I gave a 45 minute demonstration of Solver Foundation 3.0 during the Sunday afternoon session. I definitely feel like Microsoft "fits in" more than it used to. INFORMS is a great community. The title of my talk was "Introducing Solver Foundation 3.0". The demo portion of my talk featured three samples: first a mixed integer quadratic programming (MIQP) model using our Excel add-in [this link will give you the idea], and two C# models featuring integration with the nonlinear optimization (NLP) solvers built by Ziena and Frontline. The INFORMS audience is particularly savvy so they seemed to appreciate the amount of work that goes on "underneath the hood" to make the complicated look simple. In addition to demoing our new release, I wanted to leave attendees with a broader message about what our team does. For those who are already acquainted with Solver Foundation, I wanted to emphasize what I think is important about Solver Foundation: we want it to be the best way to author and solve optimization models on the Microsoft platform. This is in keeping with Microsoft Technical Computing's "modeling the world" message. I am pleased to report that the room was pretty much full and the audience seemed engaged. Questions centered on the solver plug-in features - particularly how our extensibility model is able to handle the very different capabilities of various NLP solvers. During the roundtable discussion on cloud computing I know that Solver Foundation also came up.
I had a great time visiting with attendees at our booth. I was the only guy staffing it so it was empty at times (and I had to tear it down once I left the conference). For those who popped by when I was gone – my apologies. This year we brought some nice swag that included trial versions of Excel/HPC Server/VS2010/Azure which went very quickly. I had the chance to speak with many existing users in person, as well as introduce many others to Solver Foundation and Technical Computing. These kinds of one-on-one conversations are what I like best about INFORMS. Here is a picture of the booth as I was setting it up (it wasn’t quite this spartan once I was done). There were other booths besides our own – most of the solver vendors timed new releases with the conference. IBM always has a very strong presence at INFORMS. They even sponsor their own technical tracks.
The IBM booth focused not only on their very successful CPLEX solver, but more broadly on analytics (as in their commercials, which I think are quite effective).
Long time Solver Foundation partners Gurobi Optimization were in full force. Gurobi has churned out one major release per year and has truly done some impressive work. This latest release of Gurobi features QP and MIQP support, which is being featured in our own release. It was great to see the team and hear about these improvements.
SAS had a big booth featuring the new release of JMP. I like their messaging:
Frontline (who build the solver functions that ship in Excel and have several products including Solver Platform SDK) had a very interesting booth.
Frontline brought down a four node HPC cluster running their Solver Platform SDK. It is capable of sucking in spreadsheets containing solver calls and running them on the cluster. They also demoed an Azure service doing the same.
Finally, some general observations. "Analytics" seems to be very hot right now but it is perhaps old wine in a new bottle: applying math and stats in practical ways to gain insight. Operations research is analytics on steroids in the sense that it yields not just insight but decisions. It was striking to see that the career achievement awards were given to those who essentially invented analytics techniques (Fred Glover and metaheuristics) and tools (Frontline and Excel Solver) back in the 80s. The spring INFORMS Practice conference is being re-titled to explicitly include analytics, so it will be interesting to see to what extent this influences research and product offerings presented at future INFORMS conferences. The open source effort (COIN-OR) is hot as well. They had a separate dinner and interest group, and a separate technical track, including one by Lou Hafer who talked about integrating Solver Foundation and COIN! Python also remains hot. Several vendors featured their python shells during their demos, and there were several talks on the Pyomo modeling framework.
Austin was cool but regrettably I had very little time to enjoy it in an appropriate fashion. I had to pick my spots. Here is a photo from the night before my demo.
My INFORMS experience was brief but energizing. It was a lot of fun to be able to talk with our customers, partners, and people who have never even heard about what I do. It was a privilege to represent my team and show off what we have done. Check back tomorrow for news about Solver Foundation 3.0!