Hidden Activations are not Enough I: Knowledge Matrices as Higher Representations
cs.LG, cs.NE
Submitted: 2026-09-28
Updated: 2026-09-28
Comments: 86 pages, main text 35 pages, appendices 51 pages
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- Sanity Checks for Saliency Maps
- Git Re-Basin: Merging Models modulo Permutation Symmetries
- Towards better understanding of gradient-based attribution methods for Deep Neural Networks
- The Representation Theory of Neural Networks
- Double framed moduli spaces of quiver representations
- Neural Teleportation
- Mad Max: Affine Spline Insights into Deep Learning
- Revisiting Model Stitching to Compare Neural Representations
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
- Exact and Consistent Interpretation for Piecewise Linear Neural Networks: A Closed Form Solution
- Gauge Equivariant Convolutional Networks and the Icosahedral CNN
- Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance
- Deconfounded Representation Similarity for Comparison of Neural Networks
- Reliability of CKA as a Similarity Measure in Deep Learning
- Grounding Representation Similarity with Statistical Testing
- Explanations can be manipulated and geometry is to blame
- The Role of Permutation Invariance in Linear Mode Connectivity of Neural Networks
- Scaleable input gradient regularization for adversarial robustness
- On the fibers and semi-algebraicity of ReLU neuromanifolds
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks