Congratulations to SueYeon Chung on being awarded the Klingenstein-Simons Fellowship Award in Neuroscience! SueYeon Chung is an associate research scientist & project leader at Flatiron CCN - and an assistant professor NYU Center for Neural Science. Read more here: simonsfoundation.org/2023/07/05/neu…

How can you mitigate double-descent without relying on a task-tuned regularization? In my new preprint with Cengiz Pehlevan, we show that ensembling over models of different size does the trick. (1/n) arxiv.org/abs/2307.03176

🎉Happy to share an exciting new theory paper our group, just published in Physical Review Letters with Editor's suggestion📖. Also featuring in Physics Magazine. Congratulations Albert Wakhloo on this milestone! Title: Linear Classification of Neural Manifolds with Correlated Variability

Today I am thrilled to embark on a journey as an assistant professor of physics. I just sat down in my office, took a deep breath, and sketched a long-term work plan. It feels great to have bold dreams full of opportunities and challenges. Let the good times roll.

How do the spectral properties of a model influence neural prediction benchmarks? Check out our *new* paper, “A Spectral Theory of Neural Prediction and Alignment,” accepted to #NeurIPS2023 as a spotlight! arxiv.org/abs/2309.12821 w/ Abdulkadir Canatar SueYeon Chung Albert Wakhloo 🧵1/11

Wouldn't it be great if we can reliably figure out why some ANNs predict neural data better than others? When models show similar neural predictivity, how can we differentiate between them? We show a practical & theoretical soln to this, using geometry and gen error theory.

A biologically plausible neural network for online whitening with slow (synaptic plasticity) and fast (gain control) learning accepted as a spotlight at #NeurIPS2023! A collaboration with the Simoncelli group led by the brilliant Lyndon Duong, David Lipshutz arxiv.org/abs/2308.13633

Hot off the presses: ResNet hyperparameter transfer across depth and width! Tl;dr transfer for LR+schedules, momentum, L2 reg., etc. for wide ResNets and ViTs, with and without Batch/LayerNorm w/ Lorenzo Noci Mufan (Bill) Li Boris Hanin Cengiz Pehlevan arxiv.org/abs/2309.16620

Check out my most recent paper on the origins of homochirality—just published in Nature Communications today! #homochirality #chirality This paper marks the conclusion of a four-part series on this subject. nature.com/articles/s4146…

A great piece by Simons Foundation about our theory on the capacity of neural manifolds with correlated variability, led by Albert Wakhloo. Many thanks Mara Johnson-Groh & Lucy Reading-Ikkanda.

Very happy to share this work in NeurIPS 2023 with Nikhil Vyas, Blake Bordelon ☕️🧪👨‍💻, Sab Sainathan, @DepenKenpachi, and Cengiz Pehlevan on the consistent behavior of feature-learning networks across large widths arxiv.org/abs/2305.18411. What is large width consistency? Read on! 1/n

At #NeurIPS2023? Interested in brains, neural networks, and geometry? Come by our **Spotlight Poster** Tuesday @ 5:15PM (#1914) on A Spectral Theory of Neural Prediction and Alignment. w/ Abdulkadir Canatar SueYeon Chung Albert Wakhloo

Short thread about our work at #NeurIPS2023. Topics include: - representational similarity - high-D covariance estimates - noise correlations / stochastic representation - optimal transport - scientific applications of deep nets to audio data + social neuro Details below👇

Just arrived in New Orleans for #NeurIPS2023 this week. Topics I am excited about: neuro-AI, neural manifolds (representation geometry), stat physics for machine learning, interpretability, relational & causal representations My group is presenting their awesome work👇 (1/n)

👋Check out the latest findings from our lab at #COSYNE24 CosyneMeeting Thread & links to preprints to follow (1/n)🧵👇

🔥Lots of new theories on day 3 of Cosyne: [3-100] capacity for nonlinear classification of manifolds [3-105] theory of multitask learning (optimal repr geometry + geometric measures for data analysis) [3-167] tuning diversity shapes efficient representation geometry 🧵👇🏻

[1/n] Thrilled that this project with Jacob Zavatone-Veth and @cpehlevan is finally out! Our group has spent a lot of time studying high dimensional regression and its connections to scaling laws. All our results follow easily from a single central theorem 🧵 arxiv.org/abs/2405.00592

It is with great sadness that the Simons Foundation announces the death of its co-founder and chair emeritus, James Harris Simons. Jim was an award-winning mathematician, a legendary investor and a generous philanthropist. simonsfoundation.org/2024/05/10/sim…

We mourn the loss of our friend and founding benefactor, Jim Simons. Jim was visionary, brilliant, and generous beyond measure. He has left an indelible mark on our field.

NYU-CDS article about MMCR (Maximum Manifold Capacity Representations): The key principle of MMCR is to maximize the number of image manifolds (generated by nuisance variations) that can be linearly decoded in the representations, hence maximizing the efficiency of the