Research @ Gatech

Hello, After a long period of silence, I'm back to give some news. During the past three years, I finished my Master and started a thesis. I've just created a page on the researches I conducted during my stay at GeorgiaTech. More should come on my thesis soon.

CS 6601

Covered topics

In the course, we completed 8 assignments on the foundations of AI, after reading the relevant material in the textbook:

Artificial Intelligence: A Modern Approach (3rd Edition) by Stuart Russell and Peter Norvig

The course was divided in small sections on:

• Tree/Graph Search algorithms. Especially, we reviewed depth-first, bread-first, bi-directional search. We learned a lot about A* and various heuristics. We also explored AND-OR tree.
• Local search/Optimization: gradient based, local beam search, simulated annealing, genetic algorithm.
• Games strategies. This chapter was about AI in games, like chess, we touch on minimax algorithm and alpha-beta pruning. Tic-tac-toe has been introduced as a small example.
• First Order Logic and inference. We were introduced to the notations and basic inference.
• Planning with knowledge based agent. This section illustrated how to use first order logic to make plan.
• Probabilistic reasoning. We covered Bayes nets and hidden Markov models. The Kalman filter has been introduced as a small example. We also covered exact and approximate inference and inference with factor graph.
• Natural language processing has been introduced as an application of Markov Chains. We learn when to use Viterbi or forward-backward algorithms.
• Markov decision processes. More exactly, we reviewed the basic of reinforcement learning (policy iteration and value iteration).
• Machine learning. We went through SVM, logit/probit regression, decision trees, neural networks and bayesian learning.

My personal opinion

From the reading, I learned a lot on logic and reasoning for knowledge-based agent. I was already introduced to the other fields, especially to Markov decision process, Natural language processing, and probabilistic reasoning. I ever thought that first order logic was too limited to deal with real noisy sensors. But I was not aware of the powerful capacity to make plans in difficult problems.

That explains why I loved the assignment on Markov Logic Network (MLN) which proposes a bridge to bring together the cons of probabilities and first order logic. Even if the paper was a bit obscure at the beginning, I now have a good understanding of MLN due to the assignment. I think that trying to apply MLN on an easy example is the best way to go over the theory and understand why it works and why it’s a great idea. The only drawback in the assignment was that the example was too simple to express the real power of MLNs.

Overall, I felt assignments to be too calculative. I often made small mistakes that have severely decreased my grades. It would be more interesting to get some coding exercises. Coding is a good way to understand how a method works avoiding a lot of handmade calculations. I also felt the lecture was sometimes no enough theoretical. I would prefer fewer lectures on factor graph and more lectures on side topics, for example on belief theory. Moreover, it’s pity to work on first order logic without touching on Prolog.

However, this course was a good opportunity to read thoroughly the Artificial Intelligence: A Modern Approach book, which is the reference in AI. I already read a third of the second edition. Now, I have a good understanding of all the main chapters.  I really would like to mention that projects were very challenging and exciting experiences. Anyone interested in research should take this course. I learned how to well organize my papers and manage my projects. Take a look to the project page.

Hello world!

Me @ GaTech with a french baguette and Buzz

Hello world! For my final exam in one of my GeorgiaTech courses (CS 6601), we had the opportunity to create a portfolio. Here start the story of that blog. So, if you are or you attend to be a graduate student interested in Artificial Intelligence at GeorgiaTech, you should read this blog because I really advise you to take that course. The format of that course is really different from the others. There are no midterms or final. The course is organized around three mini research projects. Your partner and you will have three weeks to write a three-page paper that will be reviewed by the other students. In addition, you will have approximately eight assignments. Finally, the workload is OK.

This blog aims graduate students that are interested in Artificial Intelligence. You can find a commented syllabus of that course here. If you would like to know more about my projects, please take a look here.