The paper extends mechanistic interpretability of large language models by modeling concepts as low‑dimensional non‑linear manifolds rather than linear subspaces. It introduces a concept‑based alignment (CBA) score to compare these manifolds across layers and models, revealing block structures in intermediate layers, a shift from syntax‑dominated to mixed syntactic‑semantic concepts, and training‑dependent multilingual sharing. The study also shows that alignment patterns differ across model families and training stages, with adjacent stages aligning more closely than distant ones.
By Tido Specht, Elias Benedict Krey, Nils Neukirch, Nils Strodthoff
arXiv:2606. 29416v1 Announce Type: cross Abstract: Can a vision model truly see an object, or does it only fit surface-level visual cues?
By Xingyu Peng, Junran Wu, Yue Hou, Zhongliang Qiao, Jiaheng Liu, Shangzhe Li, Jichang Zhao, Wenjun Wu, Xianglong Liu, Yongxin Tong, Li Dong, Ke Xu
arXiv:2610.35857v1 Announce Type: new
Abstract: Do vision models encode the transition $2{\to}3$ the same way whether it appears in handwritten digits or in hand gestures, and do they "understand" th...
By Ankit Sonthalia, Arnas Uselis, Seong Joon Oh
arXiv:2512. 11000v2 Announce Type: replace-cross Abstract: Representations pervade our daily experience, from letters representing sounds to bit strings encoding digital files.
By Francesco L\"assig
Do vision models encode the transition $2{\to}3$ the same way whether it appears in handwritten digits or in hand gestures, and do they "understand" that clean${\to}$dirty is the same transformation r...
The paper introduces ICON decomposition, a method for auditing neural networks by decomposing the variance of intermediate representations into contributions from Independent Canonical Concepts (ICON). Unlike traditional decodability metrics, ICON isolates each concept’s influence while conditioning on all other concepts and the outcome, providing a clearer view of which concepts truly drive predictions. Experiments on simulated data and real-world skin‑cancer and neuroimaging tasks show ICON outperforms seven baseline methods and successfully distinguishes genuine shortcuts from merely correlated features.
By Roshan Prakash Rane, Marco Simnacher, Manuel Pfeuffer, Marc-Andre Schulz, Nys Tjade Siegel, Maximilian Dreyer, Frederik Pahde, Wojciech Samek, Sonja Greven, Kerstin Ritter
arXiv:2609.14151v1 Announce Type: cross
Abstract: Steering language models using a set of contrastive representations has been a canonical and computationally efficient method for controlling model b...
By Prajjwal Bhattarai, Tuka Alhanai
The paper introduces fundamental representational units (FRUs) in neural networks and proposes two evaluation criteria—fidelity and stability—to assess them. It demonstrates that threshold-activated sparse autoencoders (TSAEs) can identify FRUs with high fidelity and stability, supported by large-scale experiments. The study further shows that more stable FRUs are more monosemantic and can be combined across layers to reconstruct semantic structures that align with human conceptual organization.
By Chenhui Hu, Pengfei Cao, Kang Liu, Jun Zhao
arXiv:2512. 15134v2 Announce Type: replace-cross Abstract: A goal of interpretability is to recover disentangled representations of latent concepts (features) from the activations of neural networks.
By Aaron Mueller, Andrew Lee, Shruti Joshi, Ekdeep Singh Lubana, Dhanya Sridhar, Patrik Reizinger
The paper investigates how semantic meaning is arranged locally within large language models. By using two controlled probes—one cross‑lingual and one refinement probe—on four decoder‑only LLMs, it finds that after removing dominant variance directions, sentence pairs rank by semantic relatedness, semantic factors cluster around an invariant core with structured variation, refinement tightens neighborhoods, and sentence‑level versus token‑local readouts reveal complementary geometric patterns. These results suggest that LLM semantics form local anisotropic neighborhoods that contract when meaning is refined.
By Andr\'es Daniel Godoy Ortiz
arXiv:2605.05593v2 Announce Type: replace
Abstract: Despite the remarkable success of Multimodal Large Language Models (MLLMs) across diverse tasks, the internal mechanisms governing how they encode...
By Zehao Deng, Tianjie Ju, Zheng Wu, Liangbo He, Jun Lan, Huijia Zhu, Weiqiang Wang, Zhuosheng Zhang
arXiv:2607. 04525v1 Announce Type: cross Abstract: How concepts are represented in neural networks is a fundamental question in machine learning.
By Zhimin Hu, Lanhao Niu, Sashank Varma