Artificial Intelligence

  toucheatout  2006-03-15 19:24  AI  

Welcome to the AI section. I'll do my best to provide interesting stuff, mainly under the themes i'm enjoying. That is to say Machine learning or how to have this pieces of silicon make some clever choices and learn from their previous experiences to make even cleverer choices, Data Mining that is very tied to machine learning, only it specializes in real-life cases and includes all of the data preparation and the sordid gathering of data (it's the most time-consuming process), with an aim at finding patterns or building predictive models. Last (not least), Artificial life, an alternative approach to machine learning and problem solving based on imitation of what Nature does so well, so much better than our current computing schemes.
All these considerations best fit within Multiagent systems frameworks, peering into distributed computing and collective intelligence, complex systems, components or emergence problematics.

Personal research interests

Machine learning

How a software entity can change his behavior based on its results to achieve better performance? How to create a reliable decision-making model based on previous data? How to automatically adapt those models to new data? Different problems raise different issues, as when an answer isn't just direct, but involves non-rewarding steps.

Autonomous code

Software is most of the time human-driven. Often as in desktop typical setup (text editor, movie player, etc...), the programs do not do anything that the user didn't ask for (except for annoyances as companions). Even in the servers domain, operators usually launch, restart, update software by hand. Autonomous code, however is designed to be self-sufficient once launched (viruses for instance are a kind of autonomous code, as web spiders are). However, to cope with technical incertitudes, autonomous code needs replication so as to avoid single points of failures.

Situated computing

It is nowadays very common to have code that moves from machine to machine, and/or that is distributed across machines. The location of the software is then of interest when technical problems arise.

To sum it up, my (would-be) research direction steers towards autonomous, distributed systems. It is referred in the literature as Agents (multiagent systems or multi-agents systems - MAS).

Application of this research plan naturally involves internet or any other form of network (wireless (bluetooth, GPRS, wi-fi, micro-waves, ...), electricity network, existing IP networks, avian carriers (there's a couple RFC for that (also with quality of service) with IP datagrams :-)), serial connexions, ...). Of course the actual internet is the preferred and most useful, stable resource.

The specific application should preferrably be tied with an industrial project to allow for some sort of leverage and usefulness.