Wednesday, 2 April 2008

Weekly Report

After the last meeting, I set out to clarify some experiment details and to research available tools.

I have written summaries of two papers on evolutionary emergent language. I still do not understand a few important experiment details.

I intend to polish the summaries so that I can use them to help me implement the experiments with the tools that I have found.

I have looked around for suitable tools for implementing my experiments. I had a look at a large number of tools including, Mathematica, Matlab, Octave, Emergent and FANN.

I intend to try out FANN because it seems simpler to create a program that I can later be extend and scaled up. FANN stands for "Fast Adaptive Neural Network." FANN is a C library of neural network functions. It has binds for Mathematica, Matlab and Octave. I can use these as front ends to start myself off because I am more familiar with Matlab.

I intend to work on two other things as well. Firstly, I intend to prepare a plan that lays out my estimates of how long things will take and a rough direction of where I am heading. Secondly, I intend to continue writing out a plan for the literature survey that has haunted me for a while now. I have written bits and pieces all along but I am at a bit of a loss on how to put it all together with a meaningful narrative.

My survey began with literature on AI in general. This led to focusing on the potentials and problems of AI. This led to the area of NLP where the problems of AI are most acute and where there is significant potential benefit. Looking at problems in NLP led to looking at knowledge representation and specifically commonsense knowledge bases. Seeking a way to set out experiments and to quantify results, I looked at question answering systems and document summarisation systems.

After seeing the limited ways such programs handle the problem of context, I focused on that problem which led to looking at the problem of relevance. This led to looking at how the body contributes to discerning what is relevant. With this in hand, the question of what to do when relevance is less of an issue arose. This led to looking at language again and specifically evolutionary emergent language in a society of embodied agents.

In short, I begun by looking at AI and learned about fundamental issues. The first issue was limited learning capability. This was due to a lack of world knowledge and regress of symbol references. This led to the possible solution of knowledge bases combined with question answering systems, document summarization systems and cognitive architectures. However, this just made clearer context and relevance. This led to the possible solution of embodied AI. Embodied AI provides an approach to the fundamental issues and although embodied AI has been around for a while, there is still the problem of how to go from low level to high-level intelligence. This led to looking at the area of evolutionary emergent language in a society of embodied agents.

In summary, I have been clarifying experiment details and researching tools. I intend to finish my experiment details summary, replicate an experiment, prepare a plan and write up part of the literature survey.

No comments:

Post a Comment