Visualizing Weight spaces of Neural Nets
Deep learning has made great progress in Artificial Intelligence. The weight structure of the net captures the aggregated knowledge that the various learning algorithms have accumulated. Recently, researchers have started exploring this weight-space, to see if visualization can yield insights into the “thinking of the computer” for lack of a better term.
In this project you will analyze a few simple neural nets that are trained on standard problems, build a visualizer for the weights, and see if the structure of the weights can tell us something about how your network architecture evolves its knowledge.
Machine learning, game playing concepts, artificial intelligence, statistics
Combinatorial Optimization algorithms are a core field of artificial intelligence and operations research. Better algorithms are key to important application areas ranging from oil-refineries to web-search engines. No visualization for this topic exists yet.
Information: aske.plaat at gmail.com