Sridhar Mahadevan
Location: (Amherst, Massachusetts)
Personal Research Web Page: http://www.cs.umass.edu/~mahadeva
Keywords: Representation Discovery, Machine Learning, AI, Sequential Decision Problems
Posted on: Monday, June 1st, 2009
Broad Research Area: AI / Machine Learning / Robotics / Vision, Graphics / Visualization
Research Interests:
My research interests are broadly in artificial intelligence, decision-making, and machine learning. My current research involves the study of algorithms for discovering new representations from data and prior knowledge.
Many successful intelligent systems over the past 50 years are dependent on a carefully handcoded representation provided by a human expert. A major new challenge for AI research is automating representation discovery by developing algorthms for automatically constructing features or basis functions that reflect the nonlinear geometry of a data or state space.
My current research into representation discovery builds on harmonic analysis, a subfield of mathematics where spatial and temporal data is transformed into a frequency oriented coordinate system. I am exploring both global Fourier techniques based on diagonalization principles, such as eigenvector representations (e.g. Laplacian eigenfunctions), as well as multiscale representations, such as diffusion wavelet analysis. I am also exploring group representation theory for building compact basis functions on large “symmetric” spaces.
My students and I are exploring a wide spectrum of topics in representation discovery, from the design of new algorithms for learning and decision-making, as well as applications in 3D computer graphics, information retrieval, Markov decision processes and reinforcement learning, natural language processing, robot learning, and transfer learning.
