Priya L. Donti

* PFΔ: A Benchmark Dataset for Power Flow under Load, Generation, and Topology Variations (Ana Rivera, Anvita Bhagavathula, Alvaro Carbonero): bsky.app/profile/priy...

ICYMI, my students presented the following work earlier this week, and will still be around this weekend:

* FSNet: Feasibility-Seeking Neural Network for Constrained Optimization with Guarantees (Hoang Nguyen): bsky.app/profile/priy...

(ctd.)

Tackling Climate Change with Machine Learning NeurIPS 2025 Workshop: Tackling Climate Change with Machine Learning

On my way to #NeurIPS2025! Let me know if you're around and want to catch up :)

I’ll be at the Tackling Climate Change with ML workshop: climatechange.ai/events/neuri...

And look forward to participating in a panel at the AI for Science workshop: ai4sciencecommunity.github.io/neurips25.html

If you'll be at NeurIPS, please consider stopping by our poster! Wednesday, December 3 from 4:30-7:30pm PST

neurips.cc/virtual/2025...

We show that FSNet performs remarkably well across a wide range of problem classes, including QP, QCQP, SOCP, and AC Optimal Power Flow, achieving orders-of-magnitude speedups.

Surprisingly, in several nonconvex problems, FSNet even finds better local solutions than IPOPT!

FSNet combines a neural network with a feasibility-seeking step (constraint violation minimization) to ensure constraint satisfaction with significantly lower computational cost than projection. This general framework works for both convex and nonconvex problems, and comes with provable guarantees.

Excited to share our new NeurIPS 2025 paper: "FSNet: Feasibility-Seeking Neural Network for Constrained Optimization with Guarantees"

Paper: arxiv.org/abs/2506.00362

Code: github.com/MOSSLab-MIT/...

MIT News article: news.mit.edu/2025/faster-...

We're recruiting for the Climate Change AI core team! 🌍

Core team volunteers play a vital role in shaping CCAI’s work. Join us in our efforts to foster responsible AI for climate action by democratizing expertise and enabling effective coordination across sectors, disciplines, and geographies 💪

Power flow is the backbone of real-time grid operations, across workflows incl. contingency analysis & topology optimization. Our hope is that PFΔ accelerates the development of fast, feasible, and deployable ML models for power flow 🚀

#powerflow #datasets #machinelearning

6/6

Data generation workflow for PF∆. * Step 1: Sample load from presumed-feasible load convex set, create component outage, permute generator costs * Step 2: Run modified ACOPF with limits on PF output variables removed * Step 3: Store sample if converged. Else, adjust presumed-feasible load convex set, and re-sample.

PF provides (ctd.):

* Novel data generation pipeline implemented in Julia that builds on OPFlearn.

* User-friendly PyTorch InMemoryDataset class

5/

Plot of experimental results for all selected tasks

PFΔ provides (ctd.):

* Open-source PyTorch implementation of CANOS (originally developed for the optimal power flow problem, but now adapted for PF)

* Evaluations of several state-of-the-art models, including CANOS-PF, PFNet, and GraphNeuralSolver

4/

Tasks in PF∆ benchmark. * 1.1: Unperturbed training topology * 1.2: N-1 perturbed training topology * 1.3: N-2 perturbed training topology * 2.1: Low data efficiency (same set-up as 1.3) * 2.2: Medium data efficiency * 2.3: High data efficiency * 3.1: Fixed training grid size * 3.2: Small grid size training group * 3.3: Large grid size training group * 4.1: Training with hard power flow cases * 4.2: Training with augmented hard power flow cases * 4.3: Training only with hard power flow cases

PFΔ provides:

* 859,800 solved PF instances spanning 6 power system sizes and incl. N, N-1, & N-2 contingencies, alongside multiple evaluation tasks

* Close-to-infeasible cases near steady-state voltage stability limits that enable stress-testing of ML models under edge-cases

3/

This paper introduces a comprehensive machine learning benchmark for power flow (PF), capturing diverse variations in load, generation, and grid topology.

🎉Check it out:

Paper: arxiv.org/abs/2510.22048

Code: github.com/MOSSLab-MIT/...

Dataset: huggingface.co/datasets/pfd...

2/

Title, author list, and abstract for PF∆. Title: PFΔ: A Benchmark Dataset for Power Flow under Load, Generation, and Topology Variations Authors: Ana K. Rivera, Anvita Bhagavathula, Alvaro Carbonero, Priya Donti Abstract at: https://arxiv.org/abs/2510.22048

⚡Excited to share our work "PFΔ: A Benchmark Dataset for Power Flow under Load, Generation, & Topology Variations," to be presented as part of the #NeurIPS 2025 Datasets & Benchmarks Track

Led by my students Ana K. Rivera, Anvita Bhagavathula, & Alvaro Carbonero

1/

It was such an honor to meet and work with this incredible group of 25 changemakers from across different Amazonian countries, to provide training on AI & climate change and discuss practical pathways forward.

Resources and results of this initiative will be launched at #COP30 - stay tuned!

The images captured are actual images from cameras! AI is only used to classify which type of moth is which. For those it can't classify or is uncertain about, a human would take a look at the original camera image. So, hallucination is not a risk in this particular setting.

The Machines Finding Life That Humans Can’t See A suite of technologies are helping taxonomists speed up species identification.

"In a single week, AI processed many thousands of images each night, in which experts detected 2,000 moth species—half of them unknown to science."

Cool article on AI for large-scale biodiversity monitoring, feat. my awesome colleague @drolnick.bsky.social!

www.theatlantic.com/science/2025...

Fighting for the health of the planet with AI A profile of MIT Assistant Professor Priya Donti explores her approach to applying machine learning to optimize renewable energy.

Thank you to MIT News for the kind feature! ⚡️☀️

news.mit.edu/2025/fightin...

Trump asks 9 colleges to commit to his political agenda and get favorable access to federal money The White House is asking nine major universities to commit to President Donald Trump’s political priorities in exchange for more favorable access to federal money.

Context on the email sent to 9 universities, including MIT: apnews.com/article/trum...

Text of the Compact: www.washingtonexaminer.com/wp-content/u...

MIT on the USG's Compact promising preferential treatment in exchange for specific on-campus changes:

"Fundamentally, the premise of the document is inconsistent with our core belief that scientific funding should be based on scientific merit alone"

orgchart.mit.edu/letters/rega... 💪

TIME100 AI 2025: Priya Donti Find out why Priya Donti made TIME’s list of the most influential people in artificial intelligence

Link to article: time.com/collections/...

And the full list: time.com/time100ai

While my name may happen to be the one on the list, this recognition reflects the work of many. Grateful to the team at @climatechangeai.bsky.social (esp. my co-leads David Rolnick, Lynn Kaack, Maria João Sousa), MIT MOSSLab (priyadonti.com/group), and my PhD advisors (Zico Kolter, Inês Azevedo)

Priya Donti's picture in the TIME100 AI frame

Beyond humbled to be on this year's #TIME100AI.

AI can be an asset for climate & energy -- but only if its development is guided by actual climate needs & planetary limits. Shoutout to those in the community working to shape a responsible, equitable, climate-aligned AI future 🌍💪

AI Climate Academy COP 30 Applications are now open for the AI Climate Academy pilot workshop, aimed at strengthening capacities in artificial intelligence (AI) applied to climate action in the Amazon region.

We are pleased to announce the AI Climate Academy partnership!

📅Pilot workshop from Oct 13 to 17, 2025, to be held in Belém, Brazil.

🚨Call open for participants from Amazonian countries. Deadline: Aug 29, 2025: itsrio2.typeform.com/AIClimate

Authors: @emarche.bsky.social, Benjamin Donnot, Constance Crozier (@gtresearchnews.bsky.social), Ian Dytham (@neso-energy.bsky.social‬), Christian Merz, Lars Schewe (‪@edinburgh-uni.bsky.social‬), Nico Westerbeck, @cathywu.bsky.social‬, Antoine Marot, ‪Priya Donti

Excited to share RL2Grid, an RL benchmark for power grid operations ⚡

We present tasks and environments for grid topology optimization, an important challenge in power systems that encompasses a number of open research questions in RL.

Check it out! More details below ⬇️

Congrats Alex!!

Our paper on detecting abandoned oil wells with machine learning, led by @pratinavseth.bsky.social, is accepted at ICML 2025! These wells are a major source of emissions (and groundwater pollution).

More details in the thread, and preprint here: arxiv.org/abs/2410.09032

Should LLMs be used to review papers? AAAI is piloting LLM-generated reviews this year. I wrote a blog post arguing that using LLMs as reviewers can have bad downstream consequences for science by centralizing judgments about what constitutes good research.

bryanwilder.github.io/files/llmrev...

I started to put together a starter pack for research in AI+Ecology, check it out and let me know if you would like to be added!

go.bsky.app/8zugFF6