Sriram Kumar

I am a computer vision engineer at Perceptimed Inc., where I work on computer vision and machine learning. My current work involves utilizing deep learning models for image classification task.

During my internship at Perceptimed Inc., I developed pipelines for automated tuning and validation of machine learning algorithms using Bayesian optimization schemes. I also worked on feature engineering and dimensionality reduction methods for high dimensinoal data visualization.

I completed my Masters in Electrical ENgineering at Rochester Institute of Technology, where I was advised by Andreas Savakis. My masters thesis addresses the problem of robust visual domain adaptation and image set classification in the context of manifold learning. I've spent a summer at BSIA working on discriminative dictionaries for robust signal classification. I did my bachelors at the Anna University.

Email  /  CV  /  Thesis  /  Google Scholar  /  LinkedIn  / IPython notebooks

Research

I'm interested in computer vision, machine learning, statistics, optimization and artificial intelligence. Much of my research lies in modelling the underlying manifold of images.

Grass_PA_new

Learning a Perceptual Manifold for Image Set Classification
Sriram Kumar, Andreas Savakis
International Conference on Image Processing (ICIP), 2016
poster / bibtex

DAflow

Robust Domain Adaptation on the L1-Grassmannian Manifold
Sriram Kumar, Andreas Savakis
Computer Vision and Pattern Recognition Workshops (CVPRW), 2016
poster / bibtex

DLpaper

Joint and Discriminative Dictionary Learning for Facial Expression Recognition
Sriram Kumar, Behnaz Ghoraani, Andreas Savakis
Electronic Imaging, 2016
slides / bibtex

Course Projects
gc Object recognition using Grassmann manifold
Sriram Kumar
Machine Intelligence (CMPE - 789), Fall 2014

Modelled multi-view images as low dimensinoal manifold and perform object recognition. Here we explore Grassmann manifold which is collectino of linear subspaces.

pca Mathematical modelling of a single neuron to perform Principal Components Analysis
Sriram Kumar
Adaptive Signal Processing (EEEE - 768), Spring 2014

The goal of this project was to learn principal components from a dataset using Hebbian learning. We used Oja and Sanger learning to extract single and multiple principal components.

svm_pic Facial Expression Recognition
Sriram Kumar
AI Explorations (EEEE - 647), Spring 2014

This project involved developing a facial expression recognition system. We used histogram of oriented gradients (HOG) as our features and used support vecctor machine (SVM) to perform the classification. We also expreimented with scale invariant feature transform (SIFT) which was densely computed. We used nearest neighbor as our baseline.

stock_preview

Statistical analysis of stock data
Sriram Kumar
Matrix Methods (EEEE - 603), Fall 2013

This project involved an exploratory data analysis on stock data. This included predictive modelling, analyzing trends, visualzing the spectrum using Fourier analysis.

Master's Thesis
thesis_preview

Learning Robust and Discriminative Manifold Representations for Pattern Recognition
Sriram Kumar
Rochester Institute of Technology, 2016
bibtex

Teaching Assistant
RITLogo

CMPE - 380 Applied Programming

CMPE - 610 Analytical Topics for Computer Engineering

EEEE - 678 Digital Signal Processing


This guy...