Rodrigo Alvarez Icaza - Research Experience

Growing Connections: Explorations in Computational Neuroscience
The idea behind this project is to explore biologically inspired mechanisms of development and learning like neurogenesis and neuroplasticity and implement them in a computational model applicable to robot control. A long term goal is the creation of an ASIC through VLSI. Click on the above link to view an article, in submission for publication, about the current advances.

This was my undergraduate thesis for the B.S. in Mechanical and Electrical Engineering which earned an Achievement of Excellence Award. Click on the above link for an overview of the design and its processing architecture.

Walking Machine -1997
The design of a biped robot capable of walking with the fewest degrees of freedom possible under the direction of Dr. Jaime Gómez. Click on the above link for a description of the mechanical design process and the control outline for the robot.

VLSI implementation of Neural Networks - 2001
This was my introduction to neural networks and my first conscious attempt at AI which later evolved into my current computational neuroscience project.
It began as a naive and uniformed attempt to create a massively parallel VLSI chip dedicated to the processing of neural networks. As I studied the traditional models (perceptron, backpropagation, Kohonen, etc.) I mapped the mathematical processing necessary for each signal as it propagated through the network during operation and training with hopes of later implementing this topology with logic gates or state machines.
I soon realized that even if each node was made small enough so that a large number of them could fit in a single chip, the limiting factor would be the interconnectivity.
Despite my disappointment with traditional Neural Networks this project helped me understand the challenges to overcome with computational neuroscience and the type of computations that can be expected from a neuron and a network.

Pente-1996
Originally this was a Structured-Programing final project implementing the ancient Japanese board game of Pente. The finished project was so complete that I decided to try my luck as a shareware so I translated the interface to english, improved the computer opponent and uploaded it. Its success was amazing! I received registrations from all around the globe including a few from Japan. It was also published on several CDs in the US and Europe.
At the time I had no idea that computer opponents were considered AI, or that there were books on the subject, so in a sense it was a remarkable achievement to come up with rule-based and adversary-search algorithms on my own...
Feel free to download a copy and try to beat it. It will run on any Mac. Download (408KB)

Gravity:Computer Simulation of Big Bang -1995
For some reason I had become suspicious of the Big Bang theory. I just could not visualize how a radial explosion would eventually cause everything in the universe to end up spinning around its axis or some other body. My curiosity grew so strong that I decided to program a little simulation.
Even though gravity is the weakest force in physics, and without wanting to overcomplicate the program, I decided to use Sir Isac Newton's theories on gravitation as the basis for my sim. I created a set of particles with random mass arranged along the surface of a small sphere and gave them an initial radial velocity with random magnitude. Their velocities/trajectories would then be influenced by the gravitational attraction of every other particle in accordance with classic mechanics.
Initially I tried assigning a very small mass to each particle and using the real gravitational constant but my old CPU could not handel such small influences so I had to increase overall mass and the gravitational constant until an attractive behavior emerged. This produced a lot of slingshots and soon all particles would be launched very fast and very far but would not form spinning systems. The solution was to implement a collision scheme that simply conserved momentum and merged particles when they were too close.
Finally things started to spin around each other forming systems but they were hard to track as they flew away. Adjusting the graphical interface to be able to set a particle as a fixed frame of reference produced very interesting results.
Not only did complex, although unstable, systems formed but I discovered that textbook perfect elliptical orbits are unreal. Actual planet orbits are very wobbly ellipses(proportion in mind) caused by the influence of moons. Even the sun's position wobbles by the influence of planets which is exactly what modern astronomers use to detect distant planets.
In conclusion, although this basic simulation was astronomically far from accounting for the amount of mass/energy, number of particles and phenomena involved in the Big Bang, I was convinced that a radial explosion can lead to for the formation of rotating systems.
Click here for some movies.

Solving the Chinese Solitaire - 1994
This ancient game consists of a board with 33 slots each with one marble as shown in the figure. The goal is to start with the center marble removed and capture the rest of them by jumping them. A marble can only jump over one marble to an empty slot in a horizontal or vertical direction. The jumped marble is then removed.
My Xgirlfriend's father used to play this game for hours every night without success until one night he challenged me to try. I gladly accepted the challenge and promised a solution for the next day without even trying to solve it by 'hand'. When I got home it took me around 2 hours to program a simple depth-first net search algorithm capturing the game's layout and rules. Surprisingly it only took about 15 minutes (at 25MHz) to find a perfect solution!
Even more surprisingly the challenger accused me of cheating by using a computer, he claimed that anyone could use a computer to solve it... ???
Click here for the solution.