Gautam Kowshik - Homepage

Research

Computational Behaviour Group GAMBIT (Geospatial Analysis of Motion-Based Intelligence and Tracking): Movement data can be combined with geospatial information and transformed into probabilistic graphical models that represent both social and temporal relationships between objects in the observed area. We then apply machine-learning techniques to cluster patterns in these graphical models to assist human users in performing strategic level analysis such as behavior prediction and anomaly detection. We recently published a paper[_] to International Conference on Machine Learning (2013) on a new class of classification problems in Machine Learning.

Computational Behaviour Group Spatiotemporal Patterns in Basketball: New technologies deployed in the NBA enable automatic collection of fine-grained spatio-temporal data that tracks player and ball movements, opening up a wide range of pattern recognition and sports analysis problems. We develop and apply machine learning, data analysis, and visualization tools to both validate long-held beliefs and uncover the hidden truths behind the game.; Click This Video (takes 5-10 secs to load)

Alternate Reality Gaming TeamIt: This project investigates the problem of collaboration and coordination among multiple teams and individuals, a problem which arises in many domains including disaster response. The TrackIt tool enables real-time situational awareness and task coordination among many individuals and teams, and the TeamIt Alternate Reality Game (ARG) enables team-based human experiments that run on iPads/iPhones and test real-world scenarios out in the field. Check out our project page.

Business Intelligence BI Team @ InMobi Inc.: A data ETL (Extraction-Transform-Load) and Analytics platform that allows to store and query terabyte-scale data. This platform allows the company's research, product and business teams to get intelligent insights from historical data and helps in making future decisions.

Yahoo! Inc. Hadoop Grid Data Management: Yahoo! generates almost three peta-bytes of data every month in apache logs. This data is churned and massaged into semi-structured data streams that is stored and kept accessible to downstream users across the company. These data streams come from a variety of different data sources, in different sizes, different formats and demand varying service level agreements. The Hadoop Grid Data Management platform allows these data streams to be accessed, transformed and processed using the Yahoo! Grid infrastructure.

Yahoo! Inc. Grid Computing Solutions: I've adopted, evangelised and contributed to Apache Hadoop since it's inception in early 2006. At Yahoo! I contributed, maintained and benchmarked peta-byte scale clusters for their Grid Computing infrastructure. I also worked on a data processing platform, which manages data on a multiple cluster infrastructure encompassing around 40,000 servers. It acquires, transforms and validates over 100 terabytes of data every week for 2000 feeds per-day. The data is consumed and researched on by almost every team at Yahoo! including (but not limited to) Y!Search, Y!Mail and Advertising.

High Performance Computing and Simulations Fast Fluid Solver using Fluid Thin Shell Interaction Based on Immersed Boundary Method: In this course, particle and continuum simulations are used as a vehicle to learn basic elements of high performance scientific computing and visualization. This project was to showcase the learnings in the course. We implement a fast fluid solver that simulates fluid motion and garuntees stability .We present a simple approach to simulate interaction between fluid and thin shell. Instead of solely considering one way coupling, we take both forces from fluid to thin shell and from thin shell to fluid into account, which is regarded as strong coupling. We borrow idea from Immersed Boundary Method (IBM) to solve this problem. Here is a poster we presented as part of the presentation.

Machine Learning BBall Eight-Ball: The course focuses on theory, algorithms and applications of modern statistical machine learning. Topics include parametric and nonparametric methods for supervised, unsupervised learning and other paradigms. It particularly focuses are on the theoretical understanding of these methods, as well as their computational implications. We worked on a project that focused on NBA games data, aiming to shed light on the "Rebound Problem" i.e. which team will possess the ball after a missed shot has been taken. Our efforts have been on defining attributes that would aid in discriminating the data points. In this process, we encountered missing data issues which lead us to using variations of the standard classification methods and also statistical imputation to overcome this obstacle. The workflow has been a back and forth procedure of introducing features, and getting feedback from classification algorithms as we tweaked them. Here's a paper we presented for the final project.

Computer Animation and Simulation Constrained Particle System: This project implements particle systems to model a simple chain, with an emphasis on "hard" constraints. It uses the Lagrangian dynamics approach to apply suitable constraints to the particle system using Lagrange multipliers.; Click This Video (takes 5-10 secs to load)

Computer Animation and Simulation Motion Capture Interpolation: Implemented three interpolation schemes to interpolate human motion data obtained from an optical mocap system viz. Bezier interpolation using Quaternions, Linear interpolation, Spherical Linear Interpolation using Quaternions.; Click This Video (takes 5-10 secs to load)

Computer Animation and Simulation Jello Cube: A modeled jello cube that is made of a network of mass springs. The cube will stretch, contract, oscillate, change velocity, bounce off the walls of the bounding box, based on the physical laws for a mass-spring system, made up of structural, shear and bend springs. Euler integration was used to solve differential equations for the animation. The cube bounces off the wall and responds to external forces applied as a force field within the bounding box.; Click This Video (takes 5-10 secs to load)

Computer Graphics Roller Coaster Simulation: A 3D roller coaster simulation in OpenGL using Catmull-Rom splines along with OpenGL lighting and texture mapping. The simulation runs in a first-person view, allowing the user to "ride" the coaster in an immersive environment. This included creating a 3D world view around the coaster using a Sky sphere and terrain.; Click This Video (takes 5-10 secs to load)

Computer Graphics Ray Tracer: A Ray tracer that handles opaque surfaces with lighting, shadows and antialiasing. Reads in a screen description format that configures the scene objects and lighting.; Click This Image