With a few months delay, here is my three-year research agenda for 2013-2015. I have found over the years that this agenda helps me focus and avoid being distracted by too many opportunities, since my free time has shrunk a lot now that I am responsible for new (fixed-telecom) products. To make the best use of the few free week-ends and vacation days, it works best to have a clear agenda with a few challenges.
Let me start with a short feedback on the previous plan. The first topic was “Business communication model (BPCM) and Organizational architecture”. I have made good progress on BPEM & BPCM and managed to publish a paper at ICORES 2012. I have also made some progress on modeling lean management as BPO (Business Process Optimization), following Reinertsen’s tracks, but a lot remains to be done. I have made no progress on SIFOA v2, this becomes a goal for 2014.
I have published a paper on GTES at CSDM 2012 and applied GTES to Smart-Grid simulation, with success (cf. my ROADEF presentation). Still most of my research agenda on equilibriums stays for 2013-2015 (see later).
On the other hand, I have not done much on my last theme “Information Theory and Autonomous Systems”. I have dropped my lecture at Polytechnique, and new interests (lean software factories, agile methods and architecture) have displaced my previous plans regarding information systems architecture.
My new research plan is organized into three parts as usual and is clearly geared towards my next book which I intend to write in 2015, entitled “Enterprises against complexity: from forecasting to games”. I conclude from the previous retrospective that I should push a lighted agenda with fewer objectives :)
A key application of GTES has been to market equilibriums (equilibria for purists). Most of the work in the past 5 years has been with CGS (Cellular Game Simulation), a GTES model which represents the competition between phone operators. The major presentation at INRIA in 2010 was dedicated to this topic. My further work on modeling the introduction of Free into the market (cf. CSDM presentation) was based on an even simpler representation of market competition. My goal for the next two years is to deepen my analysis of “market equilibriums”:
GTES versus classical equilibriums such as Cournot.
I want to calibrate the results obtained through GTES simulation against the classical equilibrium from game theory applied to economy (including Bertrand, Stackelberg and others). The goal is two-folds: increase the credibility of GTES by reproducing results from analytical methods with simple market equations, and enrich our set of working market model with insights from economy theory.
Repeated Tenders Market Shares.
This is a new project of mine for 2013, based on a discussion with Benoit Rottembourg. The goal is to characterize equilibrium in a closed B2B market, such as repeated tenders. One may think of an IS division which is issuing a number of tenders for software development projects to the same set of ISV. Although a crude reasoning might make one think that the stronger player will take all, game theory shows that there is need to avoid monopoly which creates a more diverse situation (as one may observe in real life).
Bids, such as LTE bids in 2011.
GTES was applied at the end of 2011 to help simulate a few bidding strategies when an auction was proposed by ARCEP for LTE frequencies. The complexity of the auction, with rules that govern the bundles of frequencies allocated to bidding operators, translates into an interesting game (although it is a “one-shot” auction, with a unique bid).
GTES has proven quite successful over the past 10 years. The next step for the coming years is to turn GTES into a robust framework, so that new problems and models may be tackled with less effort and still more meaningful results. What follows is a quick overview of my goals for the next three years, most of which have been around for a while:
Include insights from relevant Game Theory communities.
GTES is by no mean an original idea, it borrows from Game Theory, Evolutionary game theory, Multi-Agent Systems, Discrete Choice Models, Stochastic Games and Markovian Decision Processes. One of my goals is to better relate GTES with its siblings and borrow from existing literature techniques to improve the stability (% of Monte-Carlo sampling that yield and equilibrium) and the convergence rate (finding this equilibrium faster). Stochastic Games (in the form of dynamic games with complete information) have shown how to incorporate learning techniques (such as Q-learning or linear reward-inaction) to converge faster towards Nash equilibriums than a simple “best response” fixed-point iterative search. I have received a lot of help from my friends at ROADEF and I am now looking at domains such as Multi-armed Bandit or online machine learning to find inspiration. I found a reference to mixing Q-learning to PHC (Policy Hill Climbing) which is very relevant to GTES
Develop better satisfaction model and heuristics.
This follows an insight gained with the S3G experiments, as well as the CGS experiments. I have used a linear satisfaction formulation which would be better replaced by a product formulation (classical insight from performance management theory). I also want to develop a set of “generic local optimization” heuristics, instead of re-writing ad hoc algorithm for each GTES instantiation (mostly a combination of dichotomist local optimization and two-opt).
Propose a cloud distributed version using a MapReduce architecture.
GTES may be one of the easiest possible algorithm to run in parallel (citer CSDM), if only because Monte-Carlo sampling is trivial to distribute. My goal is to develop a new release of CLAIRE (CLAIRE 3.4) that may run on cloud infrastructure such as Amazon ECS (lien).
The last part of my three-year-plan is a follow-up of the SIFOA work which started in 2006 and has been the main topic of my blog. I am now ready to return to modeling and computational simulation, leveraging the work made both on:
BPEM : a model of enterprise value creation (cf. the previously mentioned ICORES article). One could summarize my original intent in 2005 with the equation “SIFOA = dBPEM / dORG”, that is, what is the structural effect of organizational architecture on efficiency, as explained with a formal computational model.
BPCM : a model that characterizes communication channels through their throughput, latency and fidelity (opposite of loss). Most of my work on social networks and affiliation networks has helped me define a simpler model than what I was using seven years ago.
The key insight which I want to validate (or refute !) with SIFOA is that managing communication is mostly managing a scarce time resource which is made even rarer because of the increase of complexity (hence of communication flows) and change frequency (hence work load). The goal of this second edition of SIFOA is to provide with a computational model for Lean Enterprise 2.0 to book. This third topic contains the following:
Better and simpler BPEM & BPCM models.
There are two specific goals related to my previous work: include social network & other related work + better understanding of 2.0 channels. Still the main goal is to produce a simpler model, hence to lose some detail.
Formal evidence of lean-style pull management in an uncertain world.
Although lean is so much more than “good plumbing”, I have been surprised by the strong demonstration of lean principles with the OAI simulation. This simulation was based on application integration through business process in information systems. I would like to reproduce similar results in the world of business process efficiency, following the tracks of Reinertsen. Part of it is well-known (application of queueing theory); parts of it is more subtle (Jackson networks of queues with irregular service laws) and part of it remains to be studied (importance of control & information flows w.r.t. business process efficiency)
Showcasing the necessity of 2.0 communication tools.
Another key intuition of my previous book is that large-scale companies or divisions are faced with information flows of such magnitude that they escape conventional tools and practices (they lead to the syndrome “I am always in a meeting or answering my email”). SIFOA v2 aims at running computing experiment which will demonstrate the scalability of new forms of communication.