I am a 2nd year masters student in Computer Science at UMass Amherst where I am advised by Prof. Cameron Musco. I completed my undergraduate from IIIT Hyderabad in August 2016.
I am interested in studying the theoretical foundations of machine learning. My research interests span over randomized dimensionality reduction, generative models, compressed sensing and statistical machine learning.
I spent the summer of 2020 as an Applied Science intern at Amazon Alexa NLU where I was advised by Dr. Abdalghani Abujabal and Dr. Claudio Delli Bovi. Prior to my masters, I spent 2 years building statistical models at the Surveillance Analytics Group, Goldman Sachs led by Dr. Mayur Thakur.
Email: agajjar ‘at’ umass ‘dot’ edu
Subspace Embeddings Under Nonlinear Transformations
Aarshvi Gajjar and Cameron Musco
Conference on Algorithmic Learning Theory (ALT 2021).
We study methods for obtaining low-distortion embeddings of low-dimensional subspaces under nonlinear transformations. We show that piecewise linear approximations of many common neural network nonlinearities can be applied to give embedding results using random embedding matrices.
When I am not working, I love listening to progressive rock music (checkout my spotify). I also enjoy playing soccer if the weather permits. Happy to chat regarding research, music or even Murakami!