Publication

Found 8 results
Author Title Type [ Year(Asc)]
Filters: Author is Akshay Rangamani  [Clear All Filters]
2024
Pinto, A., Rangamani, A. & Poggio, T. On Generalization Bounds for Neural Networks with Low Rank Layers. (2024).PDF icon CBMM-Memo-151.pdf (697.31 KB)
2023
Rangamani, A., Lindegaard, M., Galanti, T. & Poggio, T. Feature learning in deep classifiers through Intermediate Neural Collapse. (2023).PDF icon Feature_Learning_memo.pdf (2.16 MB)
Xu, M. et al. The Janus effects of SGD vs GD: high noise and low rank. (2023).PDF icon Updated with appendix showing empirically that the main results extend to deep nonlinear networks (2.95 MB)PDF icon Small updates...typos... (616.82 KB)
Xie, Y., Li, Y. & Rangamani, A. Skip Connections Increase the Capacity of Associative Memories in Variable Binding Mechanisms. (2023).PDF icon CBMM-Memo-142.pdf (1.64 MB)
2022
Banburski, A. & Rangamani, A. Neural Collapse in Deep Homogeneous Classifiers and the role of Weight Decay. IEEE International Conference on Acoustics, Speech and Signal Processing (2022).
Rangamani, A. & Xie, Y. Understanding the Role of Recurrent Connections in Assembly Calculus. (2022).PDF icon CBMM-Memo-137.pdf (1.49 MB)
2021
Xu, M. et al. Dynamics and Neural Collapse in Deep Classifiers trained with the Square Loss. (2021).PDF icon v1.0 (4.61 MB)PDF icon v1.4corrections to generalization section (5.85 MB)PDF icon v1.7Small edits (22.65 MB)
2020
Rangamani, A., Rosasco, L. & Poggio, T. For interpolating kernel machines, the minimum norm ERM solution is the most stable. (2020).PDF icon CBMM_Memo_108.pdf (1015.14 KB)PDF icon Better bound (without inequalities!) (1.03 MB)