Soumith Chintala

Building machine learning classifiers for pedestrian detection and other vision applications. Advised by Dr.Yann LeCun. This work involves discriminative classification, sparse coding and deep convolution networks. Also involves writing highly optimized programs that make use of large computing clusters using OpenMPI and optimizations involving OpenMP, SSE and CUDA(GPU). The library is open-source and can be accessed at http://eblearn.sf.net

Built state-of-the-art (when published) machine learning algorithms that can recognize google street view house numbers. Used convolutional neural networks to solve this task.

Live analysis of large unlabeled user data into different emotional metrics(including live tweets). This has various applications in sophisticated spam detection, content value metrics, automatic content aggregation etc. My approach uses convolution networks in which the words are projected into a probability space and feature vectors are automatically learned without any manual tuning.

Autonomous robot navigation and 3D scene modeling using RGBD sensors for visual odometery and 3D registration. Advised by Dr. Maneesh Singh, Research Scientist, SCR

Malignant cell classification in histopathology data using Deep Convolution networks. Achieved state-of-the-art results for texture based approaches. Advised by Dr. Claus Balhmann, Research Scientist, SCR.