We build artificial intelligence that runs as fast as nature does. It sorts through forty million proton collisions per second at the Large Hadron Collider, catches black holes merging in the LIGO data stream, and searches for new physics without being told what to look for.
A multi-university institute that brings real-time AI to high-energy physics, multi-messenger astronomy, and neuroscience.
Visit a3d3.ai → Co-founder · Board Fast ML Fast Machine Learning FoundationA worldwide community building ultrafast AI for science. It develops the open-source hls4ml and SuperSONIC tools and runs the annual Fast ML for Science conference.
Visit fastmachinelearning.org → Experimental Coordinator IAIFI NSF AI Institute for Artificial Intelligence and Fundamental InteractionsAn NSF institute that brings physics and AI together, advancing AI through physics and physics through AI.
Visit iaifi.org →Modern experiments produce far more data than anyone can store. At the LHC, more than 99.99% of collisions are thrown away within microseconds, forever. Gravitational-wave detectors stream noisy data around the clock, and a kilonova's light fades within days. The decisions that shape what we can ever learn get made in real time.
Our group works where fundamental physics meets computing. We design new AI algorithms, put them on custom hardware such as FPGAs, ASICs, and GPU clusters, and deploy them inside the world's largest experiments. Then we use them to go after the biggest open questions: the nature of the Higgs boson, the identity of dark matter, and the physics we haven't imagined yet.
The most exciting result is the one nobody predicted. We build AI that can recognize it when it happens.
We use new computing and AI to see what conventional methods miss. The same core tools carry across very different sciences, and each field sharpens them for the others.
We co-created hls4ml, which compiles neural networks onto FPGAs so they run in nanoseconds, and SONIC, which serves AI to large data centers as a service. Recent work includes KAN-based lookup-table networks, which won the FPGA '26 Best Paper Award, and learning directly on the chip.
Model-agnostic anomaly detection, contrastive embeddings, symbolic regression, and physics foundation models. We built the first AI-based anomaly search at CMS and AI agents that carry out complete physics analyses on their own.
GPU-accelerated pipelines for LIGO–Virgo–KAGRA that detect black-hole and neutron-star mergers in real time. They also clean detector noise, search for unmodeled bursts, and estimate source parameters quickly enough for public alerts.
We contributed to the Higgs discovery and to the first evidence that the Higgs couples to fermions, through its decay to τ leptons. We pioneered boosted-Higgs techniques that use jet substructure to measure H→bb̄ and H→cc̄ at the highest energies.
We search for dark matter mediators at the LHC, helped write the benchmark models the whole community uses, and are building DarkQuest, a new dark-photon experiment on Fermilab's 120 GeV proton beam.
We co-invented PUPPI, now the default pileup removal at CMS. We measure multi-point energy correlators and lead the correlator trigger for the CMS Phase-2 upgrade, which reconstructs every collision within microseconds.
Our students and postdocs take on leadership roles in large international collaborations. We also help build and run the communities and institutes that link physics to computer science.
Undergraduates in our group co-author papers, graduate students lead analyses in collaborations of thousands, and postdocs go on to start their own groups. Our alumni are now faculty at UPenn and Michigan, staff scientists at CERN and SLAC, and researchers in AI and industry.
MIT UROP students and MIT–CERN summer exchange students. Alumni have gone on to Ph.D. programs at Harvard, Stanford, Cornell, and elsewhere.
APS Tanaka Dissertation Award, EPS Best Poster, Malcolm Cotton Brown Award, and Outstanding UROP Award.
Two Buechner Prizes for Outstanding Undergraduate Advising and the MIT Teaching with Digital Technology Award.
We are always looking for curious people from physics, computer science, and electrical engineering. MIT undergraduates interested in a UROP, and prospective students and postdocs, are welcome to get in touch.