Indraneel Mukherjee, CS, Princeton, Machine Learning

Indraneel Mukherjee

I am a third year graduate student in the Machine Learning group in the Computer Science Department at Princeton University. My advisor is Rob Schapire. I did my undergraduate in mathematics at Chennai Mathematical Institute, India.

Relative Performance Guarantees for Approximate Inference in Latent Dirichlet Allocation. NIPS 2008.
I. Mukherjee, D. M. Blei.
[paper.pdf | spotlight.pdf | poster.pdf]
We prove that the improvement of collapsed mean-field inference for LDA over ordinary mean-field inference decays inversely with the length of the documents. Our work suggests one should use collapsed inference for short texts, and the more efficient mean-field inference for longer documents.
Learning with Continuous Experts using Drifting Games. ALT 2008.
I. Mukherjee, R. E. Schapire.
[paper.pdf | talk.ppt | poster.pdf ]
Complex experts predict more accurately, but are also harder to learn from. Learning from binary experts has been thoroughly studied previously. We provide a master strategy achieving tightly optimal regret bounds against the powerful class of continuous/random experts.