Computer Science and Neuroscience Collaborative Work

I continued the work from my IGERT internship as my thesis project (see below). I worked on a project that aimed to use machine learning techniques to aid in the classification of a library of neuron behavior models. My goal by the end of the year was to implement a limited classification algorithm that will be able to construct a set of broad categories to initially group and classify the models by several characteristic behavior features.

I used principal components analysis, clustering techniques, and visual analysis to find classes of behavior in the phase response curve data of our model library. THis work is detailed in my Honors Thesis. If I continued to work on this, I would hope to expand this to use more features and find a more robust behavioral classification.

Summer IGERT Work:
In the summer of 2005, I was once again accepted to the IGERT program, but chose to transfer to a computer science lab that was more in line with my major and interests. I worked collaboratively with the PI to develop data analysis solutions for a large database developed by the Marder Lab at Brandeis University. The database is a set of models of the lobster stomatogastric neurons constructed by independently varying the eight input parameters of the model to produce a set of almost 2 million models that fully represent the dynamic range of the neuron behavior.

In a project titled, “Interactive Analysis of the Structure and Organization of a Large Database of Model Neurons,” we took several measures to aid in future analysis. First we developed an algorithm for the compression of voltage traces to reduce storage space and load times of individual models. We also developed several basic pattern extraction techniques, including algorithms to automatically find the period and number of spikes per period of each model. This was part of an effort to develop a library of easily extracted patterns that could be used to apply machine learning techniques to create an auto-classification scheme for the database. Finally, we developed several software tools to assist with the exploration of small portions of the database and prepare data for further study. One tool allowed a human to look through the database and classify sets of neurons manually while another tool gave the human user an overall look at a small portion of the database allowing him to observe trends in the model. At the end of the summer I gave a presentation detailing our findings. This is where I began to learn about specific machine learning techniques and how they can be applied to real data. I also discovered a deep interest in the field of machine learning that sparked my decision to choose it as my field of graduate study.

Slides from IGERT project 2005
Handout from IGERT project 2005

A shameless plug:
Brandeis University Catalyst article about my work