This is my synthesis project with Allen Hanson and Rui Wang. More information is forthcoming.

PAS is a web-application I have been developing that allows users to classify images. It is built on top of ImageJ (for image processing and feature extraction) and Weka (for classification). More information, demos, and video tutorials can be found here. Earths oceans are a soup of living micro-organisms known as plankton. As the foundation of the food chain for marine life, plankton are also an integral component of the global carbon cycle which regulates the planet's temperature. The importance of plankton for the global ecosystem cannot be overestimated. Studying plankton is important to ecological research. For example, understanding the carbon cycle is necessary to be able to predict global climate changes. On a less global scale, studying plankton can allow marine biologists to create early warning systems for detecting harmful algal blooms in coastal waters. Applications in other fields could include ship ballast water treatment, drinking water treatment, public health, bio-terrorism defense, and industrial chemical processing.

There are several image acquisition tools such as the FlowCam and VPR (Video Plankton Recorder) that are capable of taking thousands of images in a very short period of time. Hand labeling this large quantity of images requires an enormous number of man hours. In collaboration with the Machine Learning Lab at UMass, Bigelow Lab for Ocean Sciences, and the Coastal Fisheries Institute at Louisiana State University we are automating this process by classifying plankton species from low-resolution images collected in-situ. For more information on the project and publications, click here. -->

In the first part of this research we extended the congealing framework to curve data sets (project page). More info about our current work is forthcoming.

This was an independent study I did with Paul Utgoff. We used machine learning techniques to speed up heuristic search.

More info coming soon.

Our world is populated with visual information that a sighted person makes use of daily. Unfortunately, the visually impaired are deprived of such information, which limits their mobility in unconstrained environments. To help alleviate this we are developing a wearable system that is capable of detecting and recognizing signs in natural scenes. The system is composed of a sign detector and recognizer. My research focused on the recognition phase for which I used local and meta-features for attaining a fast and accurate two-level classifier. One crucial property of the classifier is that it had a low false positive rate, since false positives come at a high cost for a visually impaired using our system.

I was funded under this project as an undergraduate and it served as a major part of my honors thesis. I have a paper in a CVPR workshop that describes my approach and the results I attained. Heres another paper at the same workshop that describes the project on a higher level. While I am still contributing to the project, it is no longer my main research domain. For more information click here.

As a freshman in college I thought robotics was the coolest thing. Andy was really nice and let me volunteer in the robotics lab. I worked with John Sweeney on his mobile robots. He was working on multi-robot coordination and control at the time and I built around nine uBots for his project. Now, Patrick Deegan took over these uBots and is converting them to balancing-bots.

My first paid job in the Robotics lab was to develop a speech interface for Dexter. I used ViaVoice for the speech recognition and Festival for the speech synthesizer. After that I developed a simple object recognition software based on color cues. The object was detected through background subtraction and the color was recognized using a maximum likelihood classifier. We estimated the probability by fitting the objects pixel values with a mixture of three Gaussians (background, object and shadow). The system performed pretty well in terms of parsing commands and recognizing objects. While developing this system I got interested in symbol grounding. Deb Roys group at MIT does some interesting work in this area.