I'm thrilled to share that I'll start as an Assistant Professor at @Columbia IEOR in July 2025!! My lab will continue to push research on AI + OR for biodiversity, planetary health, and social impact. Until then, I'll be a postdoc at @UniofOxford with @t8el and @NatureRecovery.

RT @yisongyue: Thanks to @lilyxu0, @kklmmr & the other organizers for inviting me!. It's always fun to talk about AI for Experiment Design!….

RT @2plus2make5: The US government recently flagged my scientific grant in its "woke DEI database". Many people have asked me what I will d….

PDF of advert, for sharing:

I'm hiring for a machine learning data scientist & research assistant for summer 2025!. Join me on a project on invasive species management with an innovative startup doing on-the-ground removal of environmentally destructive invasives. Paid, full-time w/ possibility to extend.

@ClimateChangeAI Monday.Hall 4 #3. @ClimateChangeAI workshop, where we have:.🧪 @yisongyue keynote on AI for Science.🐘 Iadine Chades keynote on AI for Biodiversity. Panels.🌏 Opportunities for ML for Climate Action in Asia.☀️ Data-Centric ML in Climate Applications.

@ClimateChangeAI Sunday @ 9am.Peridot 201 & 206. XAI4Science workshop, where I'll be speaking on "Barriers to explainability for sustainability, and a path towards more reliable models".

@ClimateChangeAI Friday evening @ 6:30pm. @ClimateChangeAI and Future Cities Laboratory (FCL) Global, Singapore-ETH Centre satellite event. I'll be a panelist on "Data-Centric Approaches to ML for Climate Action".

@ClimateChangeAI Paper: .Deep RL + mixed integer programming to plan for restless bandits with combinatorial (NP-hard) constraints. with @brwilder @lyeskhalil @MilindTambe_AI . Poster #416 on Friday @ 3–5:30pm.

Thrilled to be here in Singapore for #ICLR2025!. Find me:. Friday @ 3–5:30pm.Poster #416 .RL with combinatorial actions. Friday @ 6:30pm.Panelist at CCAI x ETH event. Sunday @ 9am.Invited talk @ XAI4Science workshop. Monday all day.Hall 4 #3.Organizing @ClimateChangeAI workshop.

RT @juliapgjones: Two fully-funded PhD scholarships @BangorUni in conservation social science. Deadline 5th May. If you have conservation….

I hope to see many of you in Singapore at ICLR — find me at my poster (April 25 @ 3–5:30pm SGT), at the @ClimateChangeAI workshop (April 28), or please do reach out to grab a coffee!. 🇸🇬☕️🦁🏙️👩🏻‍🔬.

I'm also co-organizing the @ClimateChangeAI workshop at #ICLR2025!. Join us April 28: We're thrilled to have keynotes by @yisongyue and Iadine Chades (@MonashUni) on path-defining work on AI for science and AI for conservation. We'll be livestreaming!.

Reinforcement learning with combinatorial actions for coupled restless bandits. Paper: Code: Join our #ICLR2025 poster session!.Friday, Apr 25 @ 3–5:30pm SGT

Our SEQUOIA method significantly improves performance above baseline approaches, which are either myopic (neglecting the sequential nature of RMABs and other MDPs) or non-combinatorial (using heuristic, iterative approaches to solve the NP-hard combinatorial constraints).

Our method combines:.— DQN: deep Q-network to approximate the value function.— MILP solving: a mixed-integer program to embed combinatorial constraints over actions. The MILP solver implicitly conducts a forward pass through the neural network to compute the expected Q-value.

Sequential planning with per-timestep combinatorial (NP-hard) constraints is hard. Existing RL for combinatorial planning focus on .(a) only combinatorial state spaces, or.(b) RL as a heuristic solver for combinatorial optimization — but in single-shot (non-sequential) planning.

We focus on planning for restless bandits (RMABs), a form of constrained MDPs. RMABs are used for public health planning with @armmanindia. Traditional RMABs model only knapsack "top-K" actions, which do not capture many practical constraints such as scheduling or path-planning.

Can we use RL to plan with combinatorial constraints?. Our #ICLR2025 paper combines deep RL with mathematical programming to do so! We embed a trained Q-network into a mixed-integer program, into which we can specify NP-hard constraints. w/ @brwilder @lyeskhalil @MilindTambe_AI

RT @__justinkay: @cv4ecology is organizing a 2-week remote workshop in May to teach AI methods to former US public sector scientists - Ple….

RT @sarameghanbeery: The AI for wildlife conservation community has been impacted by recent layoffs in the US public sector; many conservat….