Efficient Temporal Reasoning

My research focuses on efficient temporal reasoning. I work on temporal graphical models, namely dynamic Bayesian networks (DBNs). More specifically, the problems I am looking at are:

  • Sparsity in large-timestep DBNs
    Any DBN (except completely disconnected ones) for a small timestep, becomes fully connected when projected over a larger timestep. However, when variables have widely separated timescales, these large timestep DBNs yield very accurate approximations.
    AISTATS 2010 paper.
  • Multiscale DBNs
    Given a micro-scale dynamic model (e.g. a cell), when can we construct a reasonably accurate yet "simple" macro-model (e.g. a tissue)? What implications does this multiscale model structure have on structure learning?
  • Handling noisy timestamps in temporal learning
    Human annotations in temporal data are often recorded with (slightly) wrong timestamps. Depending on the timescale of the relevant processes, such timestamp errors could result in significant errors in models learnt by temporal learning algorithms. It can be shown, that under reasonable assumptions, we can incorporate the noise in the timestamp without changing the complexity of inference in hidden Markov models (HMMs) and DBNs.

Previous Projects

  • SNIFF: Searching for Java code snippets using free-form queries
    Using Javadocs, we annotate a codebase with additional comments (corresponding to the methods called) and then perform normal text search along with a final intelligent code intersection step to come up with a list of relevant snippets.
    Joint work with Sudeep Juvekar and Koushik Sen.
    FASE 2009 paper.
  • A Reachability Predicate for low-level software

    Joint work with Shaz Qadeer, Shuvendu Lahiri and Zvonimir Rakamaric.
    TACAS 2007 paper and STTT 2009 paper.
  • Potential of Alternative Crops by Integration of Multisource Data
    Joint work with Anjan Sarkar, Arka Majumdar, Debapriya Chatterjee, Shibendu S. Ray and B. Kartikeyan.
    IJRS 2008 paper.
  • A priori Overload Detection and Avoidance in RT Fair Scheduled Systems Joint work with Arnab Sarkar and Partha P. Chakrabarti.
    HPC Asia 2009 paper.