arXiv:2610.17424v1 Announce Type: cross Abstract: Respecting a-priori known symmetries of underlying data is a key principle in designing efficient neural architectures. Convolutional neural networks...
arXiv:2511. 09432v2 Announce Type: replace Abstract: Machine learning (ML) models achieve remarkable performance but remain hard to interpret due to their scale and complexity.
arXiv:2604. 00316v2 Announce Type: replace-cross Abstract: Grokking occurs when a model achieves high training accuracy but generalization to unseen test points happens long after that.
CrossGMN introduces a graph metanetwork that processes a trained source network and an initialized target network simultaneously, enabling equivariant cross‑architecture weight‑space transformations. By preserving symmetry through cross‑network message passing, CrossGMN can refine target network initializations while remaining invariant to source permutations and equivariant to target permutations. Experiments demonstrate that CrossGMN accelerates knowledge distillation, transfers across datasets without retraining, and unifies compression from diverse source architectures into a common target architecture.
Equivariant Neural Networks (ENNs) have empowered numerous applications in scientific fields. Despite their remarkable capacity for representing geometric structures, ENNs suffer from degraded expressivity when processing symmetric inputs: the output representations are invariant to transformations that extend beyond the input's symmetries.
arXiv:2608. 12010v1 Announce Type: new Abstract: Equivariant Neural Networks (ENNs) have empowered numerous applications in scientific fields.
arXiv:2604. 00208v2 Announce Type: replace Abstract: Comparing internal representations is a central goal in neuroscience and machine learning, but standard linear alignment metrics (Representational Similarity Analysis, Centered Kernel Alignment, and linear regression) are frequently applied to neural activity coordinates rather than on the underlying features.
The paper investigates how a vanilla Transformer learns symmetries from finite data augmentation on point cloud datasets. It finds an ordering of learnability: non-angle-preserving symmetries are easiest, followed by angle-preserving symmetries, and finally base angle-preserving subgroups such as translation, rotation, and scale. The study also examines the Transformer's extrapolation behavior, performs a structural analysis of trained models to uncover interpretable mechanisms for invariance, and extends these findings to equivariant functions, suggesting that the identified mechanisms can serve as building blocks for learned equivariance.
arXiv:2606. 19538v1 Announce Type: new Abstract: Convolutional networks, recurrent networks, and transformers each encode different inductive biases -- locality, sequential memory, and content-dependent pairwise interaction -- and have remained mathematically distinct since their inception.
arXiv:2505.09716v3 Announce Type: replace-cross Abstract: Out-of-distribution (OOD) generalisation is considered a hallmark of human and animal intelligence. To achieve OOD through composition, a sys...
The paper introduces MATCHA, a method that learns explicit many-to-many layer maps between independently trained language models. MATCHA decomposes the alignment into a layer map—an explicit target‑by‑source matrix—and a shared feature map, allowing joint learning of layer correspondence and feature transformation from prompts. Experiments on 42 model pairs across three families show that MATCHA reconstructs target activations more faithfully and improves retrieval metrics, while enabling transfer of activation‑space interventions between models.
arXiv:2604. 11613v4 Announce Type: replace-cross Abstract: Transformers can perform in-context classification from a few labeled examples, yet the inference-time algorithm remains opaque.