Hello!#
I am a postdoctoral research fellow at Harvard University, specializing in representation learning methods for decision-making. My research focuses on developing representations that support counterfactual reasoning, generalize across datasets and tasks, and integrate external commonsense knowledge. This involves integrating advancements from reinforcement learning, foundation models, and statistical techniques. My work also includes representation learning for structured domains such as images, graphs, spatiotemporal data, and higher-order topological data. Additionally, I have collaborated on various AI applications for social impact, particularly in public health and climate change mitigation.
I earned my Ph.D. in Statistics from the University of Texas at Austin, focusing on reinforcement learning, computer vision, and statistical applications. Prior to my Ph.D., I completed a B.S. in Applied Mathematics at ITAM and an M.S. in Pure Mathematics at the University of Cambridge. I have held internships at Meta AI (FAIR) and Intel AI. I was also a member of the UT Austin Villa Robot Soccer Team, where I developed a deep-learning vision system for autonomous soccer robots.
Below is a photo of our preparation for a match at the 2022 Robocup competition in Bangkok:
Research#
My research interests focus on the following key areas:
Representations for counterfactual reasoning and planning: Developing representations that provide reliable guarantees for counterfactual reasoning, particularly in offline settings. This research intersects with causal inference, offline reinforcement learning, and world models.
Self-supervised learning: Developing learning methods from unlabeled or partially labeled data across multiple modalities and tasks. The goal is to create foundation models that can be easily fine-tuned for counterfactual inference and sequential decision-making tasks, robust to distributional shifts, and adaptable to new features or datasets.
Common-sense and external knowledge: Fascinated by AI systems that can leverage external common-sense reasoning to overcome tabula rasa learning. With the advent of LLMs and foundation models, combining LLMs and other foundation models with learning embeddings of features to improve the generalization of models to new tasks and domains.
Representations in structured graph-like domains: Learning representations in structured domains such as graph-based and temporal data. In addition to graph neural networks, I have adopted the topological deep learning paradigm, which allows for capturing higher-order interactions in the data.
My work has been published in top-tier conferences such as NeurIPS, ICLR, and AAAI, and in scientific journals such as the Proceedings of the National Academy of Sciences. Supported by the National Science Foundation and the National Institutes of Health, this research has significant implications for public health, particularly in designing AI-driven interventions and policies for climate change adaptation and mitigation.
⚠️ Our group is looking for motivated students, interns, and postdocs to join our research efforts. If you are interested in working with us, please reach out.
News#
[2024-06-15] ✨ New grant as co-PI funded by the NSF and NIH for “Synergizing Topological Deep Learning and Spatio-Temporal Causal Inference.” This methodological grant will enable new possibilities for (possibly multi-modal, multi-resolution) spatiotemporal data.
[2024-05-22] 📃 New ArXiV paper: E(n)-Equivariant Topological Neural Networks. Check out our blog post.
[2024-05-17] 🔥 Our paper Causal Estimation of Exposure Shifts with Neural Networks has been accepted to KDD 2024. See you in Barcelona!
[2024-02-01] 🔨 Our workshop Training Agents with Foundation Models in the Reinforcement Learning Conference (RLC) 2024 is to be held on August 9th, 2024. We will release the website and call for papers soon! Reach out to
tafm.rlc@gmail.com
.[2024-02-01] 📃 New manuscript: Optimizing Heat Alert Issuance with Reinforcement Learning. My first paper as senior author.
[2024-01-16] 🔥 New paper: SpaCE paper has been accepted to ICLR 2024. See you in Vienna!
[2024-01-15] 🤖 Created a LIVE CV, powered by retrieval augmented generation. Have fun asking your own questions about my research and work experience.
[2023-12-10] Started this new website based on Chris Holdgraf’s new blog template.
[2023-09-01] Promoted to Research Associate at Harvard University. I will continue my research as usual but take on more projects in a senior role and propose grant applications.
[2023-05-15] ✨ New grant as Co-PI awarded by the Harvard Chan-NIEHS to develop new computer vision architectures that are robust for prediction under covariate shift (with applications to projecting climate change’s health impacts).
[2023-03-01] 📃 New ArXiV paper [Causal Estimation of Exposure Shifts with Neural networks](https://arxiv.org/pdf/2302.02560.p df).
[2022-08-15] Started a postdoc at Harvard University, Department of Biostatistics.
CV#
Try my 🤖 Live CV Chatbot here, powered by LLMs and RAG. Have fun asking your own questions. You can also download an outdated pdf ⬇ here. Last updated: 2024-03-30.