Falcon Dai

moon indicating dark mode
sun indicating light mode

I like to explore and create. I spend most of my time studying how to realize artificial intelligence in physical and virtual environments. I graduated from the University of Chicago with B.S. in mathematics and B.A. in physics (both with honors), PhD in computer science (specializing in reinforcement learning) at TTI-Chicago advised by Professor Matthew Walter. I currently work at Symbolica AI as a researcher.[my CV]

Research interests

  • Reinforcement learning (optimism, regret, safety)[2024a][2021a][2019b]
  • Meaning and interpretation (semiotics, semantics, metaphors)[2019e]
  • Social implications of machine learning technologies (algorithmic safeguards, human-centric objectives)[2019a]
  • 3D computer vision (surface normals, geometric landmarks)[2019d][2016a]
  • Language modeling (Chinese, glyph, summarization)[2024a][2019a][2018][2017]
  • Robotics (reinforcement learning, imitation learning, tactile sensing)[2024a][2024b]
  • Human cortical processing of language (electrocorticography, language mapping)[2020][2016a]

Publications

  • Yunis D, Jung J, -, Walter MR. Subwords as Skills: Tokenization for Sparse-Reward RL. Neural Information Processing Systems (NeurIPS), 2024.[preprint]
  • -, Walter MR. Loop Estimator for Discounted Values in Markov Reward Processes. Proceedings of Association for the Advancement of Artificial Intelligence Conference (AAAI), 2021.[preprint][code][talk][slides][poster]
  • Zheng W, Minama Reddy GK, -, Chandramani A, Brang D, Hunter S, Kohrman MH, Rose S, Rossi M, Tao J, Wu S, Byrne R, Frim DM, Warnke P, Towle VL. Chasing Language Through the Brain: Successive Parallel Networks. Clinical Neurophysiology, 2020.[paper]
  • -, Walter MR. Maximum Expected Hitting Cost of a Markov Decision Process and Informativeness of Rewards. Neural Information Processing Systems (NeurIPS), 2019.[paper][poster]
  • -, Cai Z. Towards Near-imperceptible Steganographic Text. Association for Computational Linguistics (ACL), 2019.[paper][demo][talk][poster][code][nominated for best paper awards]
  • Gehrmann S, -, Elder H, Rush AM. End-to-End Content and Plan Selection for Natural Language Generation. International Conference on Natural Language Generation (INLG), 2018.[paper][code]
  • -*, Cai Z*. Glyph-aware Embedding of Chinese Characters. Subword and Character level models in NLP workshop at Empirical Methods in Natural Language Processing conference (EMNLP), 2017.[paper][poster][code]
  • Towle VL, -, Zheng W, Issa N. “Mapping Cortical Function with Event-Related Electrocorticography” in Functional Mapping of the Cerebral Cortex, ed. Richard W. Byrne. (Springer, 2016), 91-104.[chapter][book]
  • Brang D, -, Zhang W, Towle VL. Registering Imaged ECoG Electrodes to Human Cortex: A Geometry-based Technique. Journal of neuroscience methods, 64-73. 2016.[paper][software]
  • Brang D, Towle VL, Suzuki S, Hillyard SA, Di Tusa S, -, Wu S, Tao J, Grabowecky M. Peripheral sounds rapidly activate visual cortex: evidence from electrocorticography. Journal of Neurophysiology, 3023-3028. 2015.[paper]
  • Towle VL, Minama Reddy GK, -, Zhang W, Brang D, Hunter S, Kohrman MH, Marcucilli CJ, Tao J, Rossi MA, Frim DM, Byrne RW. Chasing Language Through the Brain: Three Successive Parallel Networks. Society for the Neurobiology of Language Conference, 2014.

Working papers

  • Mason W, Brenken D, -, Cruz Castillo RG, St-Martin Cormier O, Sedal A. Acoustic Tactile Sensing for Mobile Robot Wheels. In submission, 2024.[preprint]
  • -. Word2vec Conjecture and A Limitative Result. In submission, 2019.[preprint]

Selected presentations

  • Yunis D, Jung J, -, Walter MR. Subwords as Skills: Tokenization for Sparse-Reward RL. Generalization in Planning workshop at NeurIPS, 2023.
  • -, Walter MR. Finite Time Analysis of Potential-based Reward Shaping. Reinforcement Learning and Decision Making conference (RLDM), 2019. [Student travel fellowship]
  • Vasiljevic I, Kolkin N, Luo R, Wang H, -, Daniele AF, Mostajabi M, Basart S, Walter MR, Shakhnarovich G. DIODE: A Dense Indoor and Outdoor DEpth Dataset. 3D Scene Understanding for Vision, Graphics, and Robotics workshop at Computer Vision and Pattern Recognition (CVPR), 2019.[preprint][website]
  • - and others at RIPL @ TTIC. Rubik’s cube solving robot. National robotics week special exhibit at the Museum of Science and Industry, 2019.
  • - and others at RIPL @ TTIC. Checkers-playing robot. National robotics week special exhibit at the Museum of Science and Industry, 2018.
  • Schaff C*, -*, Walter MR. Towards Active Imitation Learning. Learning from Demonstrations in High-Dimensional Feature Spaces workshop at Robotics: Science and Systems conference (RSS), 2017. [Student travel grant award]
  • -, Nettsheim G. Simulation and Modeling of the Anode of the Proposed Large-Area Picosecond Photo-Detector. Chicago Area Undergraduate Research Symposium, 2011.

Past works

  • Electrocorticography (ECoG) electrode MR-CT registration: software.
  • TwiThinks, a Twitter data analytics project that studies and visualizes Twitter data. Our realtime #ivoted (for Obama vs Romney) scoreboard received 20K+ hits on Election Day 2012. It was featured on MIT-CSAIL news.
  • Once ranked 164th on Kaggle.
  • Wikipedia Plus, a Chrome extension that enhances Wikipedia's user experience.
  • 3D printing my MRI-scanned brain (reconstructed via Freesurfer): photo and mesh.
  • My pre-2015 blog Substance.
  • Focus, simulating focus+depth visual effect with CSS 3D transform and CSS filter.
  • Volume Button, a 5-star Android app that replaces broken volume buttons by touch controls in the status bar accessible anywhere.

Contact

Elsewhere on the web