Plan for Mike Nelson

Current projects and research:

Tic-Tac-Toe playing Neural Network
Neural Network to learn to play Tic-Tac-Toe based upon David Fogel's doctoral dissertation. Uses natural selection of evolving neural networks playing against a near perfect programmed player to progressively build a better playing neural network. Done without programming rules of play into the network.

Real-Time Hatching
Studying Implementing real-time hatching. This is a very cool effect presented at SIGGRAPH 2001.

FFT and Deep Water Simulation
Applying the Fast Fourier Transform (FFT) to my real-time deep water simulation to boost the frame-rate and possibly the resolution as well. The simulation uses Fourier Transform of a Phillips Spectrum.

Modeling Real-Time Destruction of 3D Objects
I've also been kicking around a way to simply and effectively model real-time destruction of 3D objects in order to more realistically model the arbitrary destruction of them. Example: blowing up a support beam of a bridge allowing it to break and partially fall. My current model is very sketchy right now, but it involves mapping a simplified support structure that is affected by gravity and weight of the other support structures. Algorithm would probably be based upon real-time deformation of cloth, springs, and the like. Structure would be skinned upon the support structure. Not sure if this will be able to attain the frame rate or even the realistic deformation. Very skeptical at this point. A possibility is deconstructing a typical vertex based 3D model into it's support structures at the time of deformation in order to boost frame rate when the model is not being deformed then rebuilding the model after deformation and saving the new support structure. So the renderer can jump back and forth between the skinned support structure and the standard vertex model.

Any comments or suggestions welcome.