arXiv:2609.24379v1 Announce Type: cross
Abstract: Mechanistic interpretability of vision transformers seeks to decompose model computation into human-readable units, but learned representations entan...
By Gautam Ranka, Shubham Santosh Pandere, Aiden Dsouza
Mechanistic interpretability of vision transformers seeks to decompose model computation into human-readable units, but learned representations entangle many concepts in each neuron. Feature superposi...
The paper studies where to place task‑specific adapters in a vision transformer to balance storage growth and accuracy. Training all contiguous four‑block placements shows an inverted‑U accuracy curve, peaking at intermediate depths, while simple weight or activation metrics favor the deepest blocks. A neuroscience‑inspired method, LS‑B, uses frozen fMRI readouts of human visual areas to select blocks whose responses vary most across tasks, yielding backbone‑specific allocations that match or exceed the best placements found by search and use only 60% of the adapter storage while staying within 1.5 percentage points of full accuracy.
By Yuan Huang, Zihan Chen, Runbin Zhang, Hongwei Ding, Changzeng Fu, Shiqi Zhao
arXiv:2608. 12408v1 Announce Type: cross Abstract: Representational similarity analysis (RSA) is increasingly used to ask which learning rules give convolutional networks brain-like representations.
By Nils Leutenegger
arXiv:2607. 19973v1 Announce Type: new Abstract: AI researchers describe state-of-the-art models as one thing repeated at scale: the Transformer, wired identically for text, pixels, or speech.
By Jaeho Seol
The paper investigates how data scaling should be viewed as an allocation problem rather than a simple sample count, focusing on spatially structured data from human brain histology. By separating unique sample count, source diversity, and spatial coverage, the authors conduct 93 pretraining runs on 11.6 million image patches from 21 brains, showing that performance improves with more unique samples, broader spatial coverage, more compute, and larger models. However, at a fixed sample budget, distributing samples across multiple subjects does not yield additional benefit, indicating that inter‑subject variation impacts generalization but adding more sources does not help when the sample count is held constant.
By Christian Schiffer, Mathis Bode, Thomas Lippert, Katrin Amunts, Timo Dickscheid