My primary interests are artificial intelligence, machine learning and its applications in developing beneficial, adaptive, self-optimizing software. My research projects typically occur at the intersection of machine learning, sequential decision making and software engineering.
I’m very grateful for the opportunity to work with experts in machine learning as well as software engineering. My thesis advisor is Prof. Alan Fern and we collaborate frequently with Prof. Alex Groce.
The best way to contact me is via email: jervis.pinto@gmail.com
In large decision problems, even sophisticated search algorithms will often perform poorly. In this project, we design algorithms that learn control knowledge from past planning experience in the form of partial policies. We develop new imitation learning algorithms for learning partial policies and analyze their regret bounds. The partial policies are used to prune actions in large MDP’s during tree search which results in significantly better decision making in less time. See our paper at UAI14. (With Alan Fern)
Can ML be used to improve automated software testing? In this project, we design algorithms for improving the runtime/coverage profile of automated test harnesses. Towards that end, we are developing a a domain-specific template scripting testing language TSTL which allows a developer to succintly define a test space using a very high-level DSL. See our paper at NFM15 for an introduction to TSTL. (With Alan Fern and Alex Groce)
We developed an intuitively simple programming model for integrating reinforcement learning into modern programming languages. The resulting API enables developers who may not be reinforcement learning experts to easily write self-optimizing programs that learn over mulitiple executions. The resulting ABP library was used by networking experts to write adaptive network protocols and by automated testing experts to write new test harnesses. (With Alan Fern, Martin Erwig, Thinh Nguyen, Alex Groce, Tim Bauer and Patrick Zhu)
Our task was to automatically track players in raw footage and label those tracks with durative semantic labels. The final result of this large effort was the first end-to-end system that learned to play the game by observing raw video. (With Alan Fern and Rob Hess)
I interned with the analytics lab at HP Labs in Palo Alto in summer 2013. I developed a learning assistant for helping non-programmers compose cloud analytics workloads on the farm (i.e., next-generation cloud analytics systems).
I spent the summer of 2012 in San Diego at a financial-services startup on a challenging information extraction problem of extracting financial data from arbitrarily formatted text and image documents.
PhD, Machine Learning, Computer Science
Oregon State University
Fall 2007 - Spring 2015 [expected]
BE, Computer Engineering
Mumbai University
2003 - 2007
Alex Groce and Jervis Pinto. A little language for testing. In the 7th NASA Formal Methods Symposium (NFM), 2015. PDF
Jervis Pinto and Alan Fern. Learning Partial Policies to Speedup MDP Tree Search. In the Proceedings of the 30th Conference on Uncertainty in Artificial Intelligence (UAI), 2014. PDF
Alex Groce, Alan Fern, Martin Erwig, Jervis Pinto, Tim Bauer and Mohammad Amin Alipour. Learning-based Test Programming for Programmers. In the Proceedings of the 5th International Symposium on Leveraging Applications of Formal Methods, Verification and Validation (ISOLA), 2012. PDF
Alex Groce, Alan Fern, Jervis Pinto, Tim Bauer, Amin Alipour, Martin Erwig and Camden Lopez. Lightweight Automated Testing with Adaptation-based Programming. In the Proceedings of the 23rd IEEE International Symposium on Software Reliability Engineering (ISSRE), 2012. PDF
Pingan Zhu, Jervis Pinto, Thinh Nguyen and Alan Fern. Achieving Quality of Service with Adaptation-based Programming for Medium Access Protocols. In the Proceedings of the Global Communications Conference (GLOBECOM), 2012. PDF preprint
Tim Bauer, Martin Erwig, Jervis Pinto and Alan Fern. Faster Program Adaptation Through Reward Attribution Inference. In the Proceedings of the 11th International Conference on Generative Programming and Component Engineering (GPCE), 2012. PDF
Jervis Pinto, Alan Fern, Tim Bauer and Martin Erwig. Improving Policy Gradient Estimates with Influence Information. In the Proceedings of the 3rd Asian Conference on Machine Learning (ACML), 2011. PDF
Tim Bauer, Martin Erwig, Alan Fern and Jervis Pinto. Adaptation-based Programming in Haskell. In the Proceedings of the IFIP Working Conference on Domain-Specific Languages (DSL), 2011. PDF
Tim Bauer, Martin Erwig, Alan Fern and Jervis Pinto. Adaptation-based Programming in Java. In the Proceedings of the 20th ACM SIGPLAN Workshop on Partial Evaluation and Program Manipulation (PEPM), 2011. PDF
David Stracuzzi, Alan Fern, Kamal Ali, Rob Hess, Jervis Pinto, Nan Li, Tolga Konik and Dan Shapiro. An Application of Transfer to American Football: From Observation of Raw Video to Control in a Simulated Environment. In Artificial Intelligence Magazine (AIM), 2011. PDF preprint
Jervis Pinto, Alan Fern, Tim Bauer and Martin Erwig. Robust Learning for Adaptive Programs by Leveraging Program Structure. In the Proceedings of the 9th International Conference on Machine Learning and Applications (ICMLA), 2010. PDF
Vivek S. Borkar, Jervis Pinto and Tarun Prabhu. A New Learning Algorithm for Optimal Stopping. In Discrete Event Dynamic Systems (DEDS), 2009. PDF (Paywall)
Reviewer for AAAI ‘15, Machine Learning Journal, IJCAI ‘13