Sam Buchanan (@_sdbuchanan) / X

Sam Buchanan

379 posts

Sam Buchanan

@_sdbuchanan

prev. @Berkeley_EECS, @TTIC_Connect, PhD @EE_ColumbiaSEAS. Training efficiency, theory and practice!

Bay Area, CA

Joined March 2015

Pinned
Sam Buchanan
@_sdbuchanan
Oct 2, 2025
We wrote a book about representation learning! It’s fully open source, available and readable online, and covers everything from theoretical foundations to practical algorithms. 👷‍♂️ We’re hard at work updating the content for v2.0, and would love your feedback and contributions
142K
Sam Buchanan
@_sdbuchanan
Oct 31, 2025
Manifold Muon stabilizes large model training, but it's expensive 💰 -- requiring an inner loop solve for each update. 💡 But you can significantly accelerate it, leading to 2.3x speedup on @thinkymachines's experiment with no performance loss! Blog and 🧵below…
21K
Sam Buchanan
@_sdbuchanan
Oct 2, 2025
Replying to @_sdbuchanan
🔗 Web link (read): ma-lab-berkeley.github.io/deep-represent… 🛠️ Github link (contribute): github.com/Ma-Lab-Berkele…
5K
Sam Buchanan
@_sdbuchanan
Jun 17, 2024
At CVPR this week for our tutorial (tomorrow; all day; …024-tutorial-low-dim-models.github.io) on low-dim models and their ubiquity in deep learning for vision. Come by throughout the day to hear more, and let me know if you'd like to meet up!
Learning Deep Low-Dimensional Models from High-Dimensional Data: Fr...
From cvpr2024-tutorial-low-dim-models.github.io
10K
Sam Buchanan
@_sdbuchanan
Sep 15, 2022
I defended my PhD this summer! Thesis is online at academiccommons.columbia.edu/doi/10.7916/p2…. Since then I've started a new position @TTIC_Connect as a Research Assistant Professor. Please reach out if you're in the Chicago area and want to chat about research/deep learning/data :-)
academiccommons.columbia.edu
Deep Networks Through the Lens of Low-Dimensional Structure: Towards Mathematical and Computational...
Across scientific and engineering disciplines, the algorithmic pipeline forprocessing and understanding data increasingly revolves around deep learning, a data-driven approach to learning features...
Sam Buchanan
@_sdbuchanan
Jan 2, 2023
The 3rd Workshop on Seeking Low-Dimensionality in Neural Networks is finally kicking off tomorrow in Abu Dhabi @mbzuai at 9am GST! slowdnn-workshop.github.io We have a fantastic bunch of speakers and educational tutorials planned, and all talks will be livestreamed. Tune in!
9.5K
Sam Buchanan
@_sdbuchanan
Oct 4, 2023
Presenting today at #ICCV2023 with @brenthyi! We improve visual quality, compactness, and interpretability for MLP+low-rank-grid 3D representations with a bit of geometry. And we prove (with math!) that it works! 🔗 brentyi.github.io/tilted 📍 Foyer Sud 120 ⏰ 10:30 AM - 12:30 PM
18K
Sam Buchanan
@_sdbuchanan
Aug 26, 2025
Replying to @sirbayes and @druv_pai
Thank you @sirbayes ! The thorough feedback is greatly appreciated, and we are working on integrating it on the book GitHub github.com/Ma-Lab-Berkele…. We plan to keep growing the book as this research continues to develop, and all feedback from the community is invaluable here!
GitHub - Ma-Lab-Berkeley/deep-representation-learning-book: Learning Deep Representations of Data...
From github.com
2.4K
Sam Buchanan
@_sdbuchanan
Sep 1, 2023
These results have me pretty excited -- the architectural tweaks we've added to this model suggest a lot of interesting avenues for theory (maybe even to prove theorems). We endeavored to write up our methodology formally in this paper so others could get started here.
AK
@_akhaliq
Sep 1, 2023
Emergence of Segmentation with Minimalistic White-Box Transformers paper page: huggingface.co/papers/2308.16… Transformer-like models for vision tasks have recently proven effective for a wide range of downstream applications such as segmentation and detection. Previous works have
17K
Sam Buchanan
@_sdbuchanan
Jan 3, 2024
Yi Ma @YiMaTweets kicking off the inaugural CPAL 2024 @CPALconf ! Looking forward to a great week of talks
5K
Sam Buchanan
@_sdbuchanan
Oct 8, 2023
Mind-boggling thing seen in Paris during #ICCV2023 -- there is actually a building in the center of this photo (mirrored exterior), can you see it? An incredible trompe-l'œil, perfect for a vision conference in Paris 🥺
6.5K
Sam Buchanan
@_sdbuchanan
Dec 13, 2023
Presenting today at #NeurIPS2023! We give a framework for parsimonious representation learning in which a (peformant, scalable) transformer-like architecture, CRATE, arises naturally from unrolled optimization! 🔗 ma-lab-berkeley.github.io/CRATE/ 📍 Poster #2006 ⏰ 5 - 7 PM
White-Box Transformers via Sparse Rate Reduction
From ma-lab-berkeley.github.io
12K
Sam Buchanan
@_sdbuchanan
Oct 2, 2025
Replying to @_sdbuchanan
This book is our attempt to answer what an "intro to AI" course in 2025 should look like. The breakneck pace of progress in AI research can make it seem like theory is cut off from practice. Our book argues that the opposite is true! A solid grasp of theory can inspire
3.8K
Sam Buchanan
@_sdbuchanan
Oct 31, 2025
Replying to @_sdbuchanan
🔗Code: github.com/sdbuch/thinky-… 🔗Blog: sdbuchanan.com/blog/manifold-… Read for full details, and try it yourself!
GitHub - sdbuch/thinky-manifolds
From github.com
1.4K