Vishakh's research and projects
This page contains short descriptions of my research and projects, which are listed in rough chronological order. Links have been provided to documents that contain further information. I plan to add more detail about each project soon. If you have any comments or questions, please use my site's contact page to send me a message.
A Comparison of Ensemble Methods for Microarray Data Analysis
Machine Learning tools are increasingly being applied to analyze data from microarray experiments. These include ensemble methods where weighted votes of constructed base classifiers are used to classify data. The purpose of the project was to compare the performance of AdaBoost, bagging and BagBoost on gene expression data from the yeast cell cycle. AdaBoost was found to be more effective for the data than bagging. BagBoost offered an advantage over AdaBoost due to the combination of the benefits of both bagging and boosting.
Using Boosting To Analyze Microarray Data
The analysis of microarray data is becoming increasingly important since it has tremendous potential for aiding the understanding of genetic expression. Gene expression profiles generated by microarray experiments can be quite complex since these experiments can involve hypotheses involving entire genomes. Machine learning algorithms are used to analyze this data in a timely, automated and cost-effective way. The computer theoretic goal of the project was to study the applicability of one particular type of machine learning algorithm, called boosting, to microarray data analysis. The biological goal was to create a predictive system for the expression levels of genes at various points in the cell cycle of yeast, given a set of regulators active at a certain time and the binding sites of a particular gene.
Ensuring efficient resource allocation in multi-agent networks
Initiatives such as the Bio-networking architecture project seek to create survivable, adaptive and scalable network applications for the increasingly heterogenous and mobile internet. For this, they draw inspiration from emergent biological systems such as ant colonies. Many of these initiatives use virtual creatures, known as agents, that migrate over network platforms and provide services. An outstanding problem before is how to ensure that agent populations corresponded to user distributions and needs. I created an information-propagation mechanism for multiagent networks, called Price Propagation, to solves it. I presented this work at the Southern California Conference on Undergraduate Research and the UROP symposium at UC Irvine. A paper based on this work has been published at the undergraduate workshop of the Genetic and Evolutionary Computation Conference (GECCO 2005).
Studying the growth and formation of networks of service-providing agents
(with Nick Urrea and Dr. Tadashi Nakano)
Agent-based networks are being increasingly used to provide network services. These networks consist of software agents that interact with users and with each other to provide services such as news collection. We used referral networks to create a generic "N-Services" model for studying such network services. A paper about this will be published soon in ACM Crossroads.
Detecting putative regulatory sequences in yeast DNA
(with Joe Bertolami, Nick Urrea and Jeff Weiss)
Simple unicellular organisms such as yeast have a lot in common genetically with humans. Thus, if we can understand and cure genetic disorders in yeast, we might be able to cure the corresponding disorders in humans. For our project class in Artificial Intelligence, we wrote code to find putative regulatory sequences in yeast DNA. Our project was based upon previous work by Jacques van Helden.
Search tools for the Hydra EST database
The Hydra EST Database project is sequencing the hydra genome. I worked with Dr. Dennis Kibler to construct search tools that let hydra researchers obtain lists of ESTs according to supplied criteria. Taxonomy Search, for instance, lists ESTs that are common to the genomes of a given set of organisms. Since these organisms are drawn from various stages of evolutionary history, information about their evolution can be inferred using the search results.
Formula 1 expert system
(with Nick Urrea)
Formula 1 always provides topics for intense discussion among fans. They often debate about scenarios such as how a certain car would fare in the hands of a driver from a different team. To shed some light on these issues, we created a Formula 1 expert system for our Expert Systems class. The actual software lets users choose a track and weather conditions, along with ten combinations of chasis, engines and drivers. The user can then view simulated results and perhaps sleep tighter knowing he is right. :-)