Tomas Singliar's Patensts and Publications
Business Intelligence / Data Science / Econometrics
♥ T. Singliar, F. Moerchen: DELi: A framework for measuring customer impact of catalog changes, Amazon Machine Learning Conference, April 2015 [Amazon internal publication]
♥ N. Rose, A. Dutta, T. Singliar: A quasi-A/B technique for ASIN experiments, Amazon Machine Learning Conference, April 2015 [Amazon internal publication]
♥ T. Singliar, F. Moerchen: Quantifying impact of sourcing catalog data, Amazon Machine Learning Conference, April 2014 [Amazon internal publication]
♥ T. Singliar, D. Margineantu: Intent estimation method and system for agents of limited perception. US Patent #8,959,042. Amends USP#8,756,177.
♥ T. Singliar, D. Margineantu: Methods and system for estimating subject intent from surveillance, US Patent #8,756,177, issued June 2014
♥ T. Singliar: Monitoring the state-of-health information for components, US Patent #8,533,133, issued Sep 2013
♥ T. Singliar, D. Marginenantu: Scaling up Inverse Reinforcement Learning through Instructed Feature Construction , Snowbird Learning Workshop 2011. [pdf]
The grand scheme of things is to create models that leverage existing data being recorded on the highways to useful ends such as detecting accidents automatically. Routing decisions that you get from MapQuest and the like give you the best expected travel time, marginally. Can we do better if we condition on traffic conditions expected at the actual time of travel? How do we model and predict the "expected conditions"?
T. Singliar, M. Hauskrecht: Towards a Learning Incident Detection System;
Workshop on Machine Learning Methods for Surveillance and Event Detection
at the International Conference on Machine Learning
ICML 2006, Pittsburgh, 2006
♥ T. Singliar, M. Hauskrecht: Modeling and learning of highway traffic volumes and their interactions, Technical report TR-06-142, Computer Science Dept, University of Pittsburgh, 2006
T. Singliar, M. Hauskrecht: Learning to Detect Adverse Traffic Events from Noisily Labeled Data;
Principles and Practice of Knowledge Discovery in Databases, ECML/PKDD 2007
, Warsaw, Poland, 2007
T. Singliar, M. Hauskrecht: Modeling Highway Traffic Volumes;
European Conference on Machine Learning, ECML/PKDD 2007
, Warsaw, Poland, 2007
T. Singliar, M. Hauskrecht: Approximation Strategies for Routing in Dynamic Stochastic Networks;
ISAIM 08 -
10th International Symposium on Artificial Intelligence and Mathematics
, Ft Lauderdale, FL, 2008
Machine Learning / Data mining
T. Singliar, D. Dash: Efficient inference in persistent Dynamic Bayesian Networks;
Appeared in UAI-08.
The noisy-or component analysis model mines binary data for common causes of link appearance.
By dividing a population of computer hosts into clusters according to features putatively indicative of worm infection susceptibility, one can improve the signal-to-noise ratio. Since we don't have access to those features, we derive the clusters from network behavior patterns.
T. Singliar, D. Dash: Online Temporal Clustering for Outbreak Detection;
6th Annual Conference of the Syndromic Surveillance Society, 2007
Petri Nets are a popular formalism for specification of concurrent systems with a solid theoretical underpinning. This paper is an algebraic characterization of PN models that allow for a concept of synchronization.