Valentino Crespi

Assoc. Professor

College of Engineering,
Computer Science and
Technology (ECST)
5151 State University Drive
Los Angeles, CA 90032-8150

Office: ET-A318
Phone: (323) 343-4596
Fax: (323) 343-6672


  • 1992: Laurea in Scienze dell'Informazione (summa cum Laude), University of Milan, Milan, Italy.
  • 1997: Ph.D. in Computer Science, Universities of Milan and Turin, Italy.



  • Combinatorial Optimization, Matrix Computation and Graph Theory

    Involved in the study of the computational complexity of the Permanent function of sparse circulant matrices and of the Lovász theta function of sparse circular graphs (1994-2004).

    Established fast algorithms and exact formulas published on Linear Algebra and its Applications and on the SIAM Journal on Discrete Mathematics.

  • Distributed Control, Surveillance, Sensor Networks and UAVs

    In charge of the DARPA TASK research project (2000-2003), developed at Dartmouth College.

    Main contributions: development of a novel design methodology for provably performant distributed control systems. Applications to problems of multi-agent UAV navigation, distributed sensor registration (in a sensor network it is the problem for all the individual sensors to establish their geographic position), and tracking with networks of minimalistic sensors.

    Results published in the Journal of Autonomous Robots and in the proceedings of the World congress on Artificial Intelligence, of the International Joint Conference on Neural Networks, of the KIMAS conference, of the AAMAS conference, of the ACM SenSys conference, and of several SPIE conferences.

  • Multi-Target Tracking, Process Query Systems and Stochastic Modeling

    Among the initiators of the DHS PQS research project (2003-2008), developed at Dartmouth College and at the Institute for Security Technology Studies, Hanover, NH.

    Contributions: development of a Process Query System, a novel revolutionary software system capable of accepting process descriptions as queries and then performing standing queries and searches against databases and data streams for evidence that instances of the queried processes exist in the data. This technology was successfully applied to build one of the best intrusion-detection systems for coordinated computer attacks.

    Results on surveillance applications published in SPIE conferences. Established results on estimating the entropy rate of Hidden Markov models published on IEEE Transactions on Information Theory.

  • Trackability, Complexification and Machine Learning of Hidden Markov Models (HMMs)

    Principal Investigator of the AFOSR Engineering Awareness research project (2007-2009), developed at Dartmouth College and at Pasadena, CA.

    Goal: establish fundamental scientific results that allow to monitor environments effectively.

    Contributions: a) developed a rigorous quantitative notion of trackability of processes/behaviors which allows to determine the "complexity" of estimating state trajectories of a target process based on a discrete-time sequence of noisy "observations"; b) developed a novel algorithm to machine learn HMMs from observed data based on the non-negative matrix factorization (NMF) of higher order Markovian statistics, structurally different from the classical Baum-Welch and associated approaches; c) developed a technique to attack and defend covert channels through machine learning certain behavioral models.

    Results on Trackability Theory published on ACM Transactions on Sensor Networks (special edition, 2008), results on machine learning HMMs using the NMF published on IEEE Transactions on Information Theory (June, 2011). See the most recent work on Attacking and Defending Covert Channels and Behavioral Models.

    See also special topics classes on machine learning languages and processes offered at CSULA.

  • The Modeling Component and the CEaS-CREST center

    Current Leader of the Modeling Component of the NSF Center for Energy and Sustainability (CEaS-CREST).

Recent Talks

Recent Students

  • Gail Casburn, M.S., distributed algorithms for sensor registration in 3D, defended in 2011.
  • Natalya Shatokhina, M.S., parameter optimization for SOLiD next generation sequencers, defended in 2011.
  • Sanmit Narvekar (undergraduate), classification problems in computer security and social networks.
  • David Gilbert, B.S., machine learning Hidden Markov Models (supported by CEaS-CREST).

BINF - Multidisciplinary Minor in Bioinformatics and Computational Biology at CSULA

  • Co-PI of the NIH Center for Interdisciplinary Quantitative Analysis (CINQA) and, together with Dr. Jamil Momand, author of BINF.

  • Designer of BINF 403 "Process Estimation & Detection in Cellular Biology" (prerequisites: CS 202, BINF 400, one of the listed Statistics course).

  • Announcements for the Students
    • 12/2011: All students interested in BINF need to complete the following prerequisites by the end of Spring quarter prior to taking BINF 400 in Fall 2012: Biol 100A, Biol 100B. Furthermore at least one of the following courses in statistics is required by the end of Fall 2012: MATH 270, MATH 474, ECON 209 and BIOL 300. Remember that Biol 100A is offered in Winter 2012 and is a prerequisite for Biol 100B, whereas the statistics classes, if offered, may be taken also concurrently with BINF 400 in Fall 2012.