About
I'm a Principal AI Researcher at DataRobot. Top of mind right now: continual learning for AI agents — systems that improve on the job, retain what works, and stay dependable as the world around them shifts.
That interest is the latest turn in a fifteen-year thread: how do you train an agent to act capably in an environment it doesn't fully understand? I've chased that question through deep RL agents playing Atari from pixels, text-game agents reading and acting in language worlds, and Pareto-optimal generative AI workflows.
- nowDataRobot — Principal AI Researcher. Continual learning and reliability for AI agents.
- 2021–2023Latitude AI — Staff Scientist. Autonomy for self-driving vehicles.
- 2016–2021Microsoft Research — Researcher. RL, text-game agents, program synthesis.
- 2010–2016UT Austin — PhD in Computer Science, advised by Peter Stone.
News
- 2025syftr accepted as an oral presentation at AutoML 2025.
Selected Publications
All Publications
2025
- syftr: Pareto-Optimal Generative AIA Conway, D Dey, S Hackmann, M Hausknecht, M Schmidt, M Steadman, N VolynetsAutoML 2025 · oral
2022
- Uni[Mask]: Unified Inference in Sequential Decision ProblemsM Carroll, O Paradise, J Lin, R Georgescu, M Sun, D Bignell, S Milani, K Hofmann, M Hausknecht, A Dragan, S DevlinNeurIPS 2022 · oral
- MoCapAct: A Multi-Task Dataset for Simulated Humanoid ControlN Wagener, A Kolobov, F Frujeri, R Loynd, C Cheng, M HausknechtNeurIPS 2022 Datasets & Benchmarks · code
- Consistent Dropout for Policy Gradient Reinforcement LearningM Hausknecht, N WagenerarXiv preprint
2021
- Reading and Acting while Blindfolded: The Need for Semantics in Text Game AgentsS Yao, K Narasimhan, M HausknechtNAACL 2021
- ALFWorld: Aligning Text and Embodied Environments for Interactive LearningM Shridhar, X Yuan, M Côté, Y Bisk, A Trischler, M HausknechtICLR 2021 · webpage
2020
- Keep CALM and Explore: Language Models for Action Generation in Text-based GamesS Yao, R Rao, M Hausknecht, K NarasimhanEMNLP 2020 · code
- Working Memory GraphsR Loynd, R Fernandez, A Celikyilmaz, A Swaminathan, M HausknechtICML 2020 · code
- Learning Calibratable Policies using Programmatic Style-ConsistencyE Zhan, A Tseng, Y Yue, A Swaminathan, M HausknechtICML 2020
- Graph Constrained Reinforcement Learning for Natural Language Action SpacesP Ammanabrolu, M HausknechtICLR 2020 · code
- Interactive Fiction Games: A Colossal AdventureM Hausknecht, P Ammanabrolu, M Côté, X YuanAAAI 2020 · code
2019
- ScriptNet: Neural Static Analysis for Malicious JavaScript DetectionJ Stokes, R Agrawal, G McDonald, M HausknechtMILCOM 2019
- NAIL: A General Interactive Fiction AgentM Hausknecht, R Loynd, G Yang, A Swaminathan, JD Williamstechnical report · code
2018
- Counting to Explore and Generalize in Text-based GamesX Yuan, M Côté, A Sordoni, R Laroche, R Tachet des Combes, M Hausknecht, A TrischlerEWRL 2018
- TextWorld: A Learning Environment for Text-based GamesM Côté, A Kadar, X Yuan, B Kybartas, T Barnes, E Fine, J Moore, M Hausknecht, L Asri, M Adada, W Tay, A TrischlerIJCAI/ICML Computer Games Workshop 2018 · code
- Leveraging Grammar and Reinforcement Learning for Neural Program SynthesisR Bunel, M Hausknecht, J Devlin, R Singh, P KohliICLR 2018
2017
- Revisiting the Arcade Learning Environment: Evaluation Protocols and Open Problems for General AgentsMC Machado, MG Bellemare, E Talvitie, J Veness, M Hausknecht, M BowlingIJCAI 2017
- Neural Program Meta-InductionJ Devlin, RR Bunel, R Singh, M Hausknecht, P KohliNeurIPS 2017
2016
- Cooperation and Communication in Multiagent Deep Reinforcement LearningM HausknechtPhD thesis, UT Austin
- Half Field Offense: An Environment for Multiagent Learning and Ad Hoc TeamworkM Hausknecht, P Mupparaju, S Subramanian, S Kalyanakrishnan, P StoneAAMAS ALA Workshop 2016
- On-Policy vs. Off-Policy Updates for Deep Reinforcement LearningM Hausknecht, P StoneIJCAI Deep RL Workshop 2016
- Deep Reinforcement Learning in Parameterized Action SpaceM Hausknecht, P StoneICLR 2016 · code
- Machine Learning Capabilities of a Simulated CerebellumM Hausknecht, W Li, M Mauk, P StoneIEEE TNNLS 2016
2015
- Deep Recurrent Q-Learning for Partially Observable MDPsM Hausknecht, P StoneAAAI Fall Symposium 2015
- Beyond Short Snippets: Deep Networks for Video ClassificationJ Ng, M Hausknecht, S Vijayanarasimhan, O Vinyals, R Monga, G TodericiCVPR 2015
2014
- A Neuroevolution Approach to General Atari Game PlayingM Hausknecht, J Lehman, R Miikkulainen, P StoneIEEE TCIAIG 2014
2012
- Using a Million Cell Simulation of the Cerebellum: Network Scaling and Task GeneralityW Li, M Hausknecht, P Stone, M MaukNeural Networks 2012
- HyperNEAT-GGP: A HyperNEAT-based Atari General Game PlayerM Hausknecht, P Khandelwal, R Miikkulainen, P StoneGECCO 2012
2011
- Dynamic Lane Reversal in Traffic ManagementM Hausknecht, T Au, P Stone, D Fajardo, T WallerIEEE ITSC 2011
- Autonomous Intersection Management: Multi-Intersection OptimizationM Hausknecht, T Au, P StoneIROS 2011
2010
- Learning Powerful Kicks on the Aibo ERS-7: The Quest for a StrikerM Hausknecht, P StoneRoboCup International Symposium 2010
2009
- For Want of a Nail: How Absences Cause EventsP Wolff, A Barbey, M HausknechtJournal of Experimental Psychology: General 2009
Open Source
- microsoft/jerichoA lightweight Python interface connecting learning agents with interactive fiction games — a benchmark for language-based agents on man-made text games.
- LARG/HFOHalf Field Offense: a multiagent RoboCup soccer subtask where the offense must get past the defense to score.
- Arcade-Learning-EnvironmentThe standard platform for training agents on Atari games. I developed the first external interface allowing ALE to be used as a library.