Adrian Weller, Academic Homepage at Columbia University
August 12, 2014













I am at the final stage of my PhD in computer science, in the area of machine learning, under the supervision of Prof Tony Jebara. I submitted my thesis on "Methods for Inference in Graphical Models" and defended it in May before a committee comprising Profs Alfred Aho, Maria Chudnovsky, Amir Globerson (Hebrew University), Tony Jebara and David Sontag (NYU). Much of the thesis is based on work in the publications below. I am making minor revisions and shall release the final version soon, please contact me if interested in a preliminary draft or CV.

Most of my academic research relates to graphical models but I'm also very interested in other areas including: finance, anything on intelligence (natural or artificial), deep learning, reinforcement learning, evolution, Bayesian methods, time series analysis and methods for big data.

I have TA'd and then taught the Masters level machine learning course at Columbia (see here), which was a great experience.

I can be reached at my first name (at) cs.columbia.edu, where (at) means the @ sign.


Publications

A. Weller and T. Jebara. Approximating the Bethe partition function. In Uncertainty in Artificial Intelligence (UAI), 2014.
Poster

A. Weller, K. Tang, D. Sontag and T. Jebara. Understanding the Bethe approximation: When and how can it go wrong?. In Uncertainty in Artificial Intelligence (UAI), 2014.
Poster

K. Tang, A. Weller and T. Jebara. Network ranking with Bethe pseudomarginals. In NIPS Workshop on Discrete Optimization in Machine Learning, December 2013.

A. Weller and T. Jebara. On MAP inference by MWSS on perfect graphs. In Uncertainty in Artificial Intelligence (UAI), 2013.

A. Weller and T. Jebara. Bethe bounds and approximating the global optimum. In Artificial Intelligence and Statistics (AISTATS), 2013.

Earlier work

A. Weller, D. Ellis and T. Jebara. Structured Prediction Models for Chord Transcription of Music Audio. International Conference on Machine Learning and Applications (ICMLA), December 2009.

These methods were used to provide a slight improvement to Dan Ellis' existing, powerful approach to chord transcription, which led to us submitting the best entry to the MIREX open competition that year, see results here. If interested, a brief description of the overall 2010 LabROSA chord recognition system is given here.

Selected presentations

Oxford Department of Statistics, September 2014