Mackenzie Weygandt Mathis
@trackingactions.bsky.social
12K followers 730 following 560 posts
Scientist 👩‍🔬 & EPFL Prof 🇨🇭 | DeepLabCut.org , 🦓 cebra.ai | neuroscience & ML 🧠 mackenziemathislab.org | ✨CSO at Kinematik.ai | occasionally 🐦‍⬛birds/🌱outdoors/🍣food/👠fashion
Posts Media Videos Starter Packs
trackingactions.bsky.social
It’s going to be fabulous - I can’t wait! 🙌
trackingactions.bsky.social
Enjoying the extra flare in the office today courtesy of my lab ❤️🥹 - a little congrats that I gave my tenure talk yesterday! 🤞
Reposted by Mackenzie Weygandt Mathis
kristorpjensen.bsky.social
I’m super excited to finally put my recent work with @behrenstimb.bsky.social on bioRxiv, where we develop a new mechanistic theory of how PFC structures adaptive behaviour using attractor dynamics in space and time!

www.biorxiv.org/content/10.1...
trackingactions.bsky.social
Fall US trips booked! I'll be speaking at Scripps, SfN, and ICCV 🚀 Who will be at #ScrippsNeuroSymposium2025, #SfN2025 and/or #ICCV2025? Lots of ☀️🕶️🌴 & 🧠 in Oct/Nov!
trackingactions.bsky.social
Not just an amazing dataset, excellent 3D pose, shape, and eye tracking in the real world! Bravo @trackingskills.bsky.social et al!

PS - Exactly this kind of data we predict will be critical for long-term health monitoring @monosovlab.bsky.social et al: www.sciencedirect.com/science/arti...
Reposted by Mackenzie Weygandt Mathis
newyorker.com
A.I. tools are already shaping patient care—but we should be wary of letting them diagnose us. “Patients and doctors alike could think of A.I. not as a way to solve mysteries but as a way to gather clues,” Dhruv Khullar writes.
If A.I. Can Diagnose Patients, What Are Doctors For?
Large language models are transforming medicine—but the technology comes with side effects.
www.newyorker.com
Reposted by Mackenzie Weygandt Mathis
Reposted by Mackenzie Weygandt Mathis
mollfw.bsky.social
DeepLabCut pose estimation (@trackingactions.bsky.social) revealed high intitial motor variability, which is essenatial for motor learning (Dhawale et al., 2017, Annu. Rev. Neurosci.).
Reposted by Mackenzie Weygandt Mathis
kevinjkircher.com
New paper in Nature finds that global geologic CO2 storage potential is ~90% lower than previously believed. Many candidate sites could leak CO2 back into the air or into groundwater, undermining climate mitigation efficacy or putting nearby human health at risk.

www.nature.com/articles/s41...
A prudent planetary limit for geologic carbon storage - Nature
A risk-based, spatially explicit analysis of carbon storage in sedimentary basins establishes a prudent planetary limit of around 1,460 Gt of geological carbon storage, which requires making explicit ...
www.nature.com
trackingactions.bsky.social
I might be biased, but I think this is one of the most exciting developments in generalized musculoskeletal control in years ⬇️. Congrats to the team (I’m not involved!) 💪🔥
trackingskills.bsky.social
Something exciting just left the gym -- meet Arnold the first generalist musculoskeletal control policy!

The human body has 600+ muscles, yet we seamlessly coordinate them for everything from whispering a syllable to felling a forrest. How? www.arxiv.org/abs/2508.18066
Reposted by Mackenzie Weygandt Mathis
trackingskills.bsky.social
Something exciting just left the gym -- meet Arnold the first generalist musculoskeletal control policy!

The human body has 600+ muscles, yet we seamlessly coordinate them for everything from whispering a syllable to felling a forrest. How? www.arxiv.org/abs/2508.18066
Reposted by Mackenzie Weygandt Mathis
icepfl.bsky.social
EPFL, ETH Zurich & CSCS just released Apertus, Switzerland’s first fully open-source large language model.
Trained on 15T tokens in 1,000+ languages, it’s built for transparency, responsibility & the public good.

Read more: actu.epfl.ch/news/apertus...
trackingactions.bsky.social
This was a wonderful project to work on! It also, in part, sparked a lot!

Beyond just our handles: @trackingskills.bsky.social aka trackingplumes (Bluesky and X), when I wanted to track mouse hands & Alex noses, we teamed up in 2017 to make @deeplabcut.bsky.social 💜

Delighted to see this out! 🙏🏼
Reposted by Mackenzie Weygandt Mathis
bendichter.com
Yes, AmadeusGPT is a great example of the kinds of structure that can make these LLMs truly useful for science
trackingactions.bsky.social
Selfishly, I’ll plug our tool LLM agent in system #AmadeusGPT - which lets you vibe code your behavioral analysis ;) and even gives you interpretable and verifiable outputs!

www.mackenziemathislab.org/amadeusgpt
trackingactions.bsky.social
This approach enhances the reliability of trigger reconstruction, making it capable of distinguishing between clean & trojaned models. 🚀

Congrats to all the authors who did an amazing job! 3/4
trackingactions.bsky.social
By employing a diffusion-based generator guided by the target classifier, #DISTIL iteratively produces candidate triggers that align with the model's internal representations associated with malicious behavior. 2/4
trackingactions.bsky.social
My lab has been pushing into explainable, robust, & theoretically-tractable AI models for science 💪

New! Accepted to #ICCV2025 we introduce #DISTIL - led by amazing PhD student @mirzious.bsky.social - we propose a trigger-inversion method for DNNs that reconstructs malicious backdoor triggers 1/4
Reposted by Mackenzie Weygandt Mathis