Mechanistic mode connectivity

ES Lubana, EJ Bigelow, RP Dick… - International …, 2023 - proceedings.mlr.press
We study neural network loss landscapes through the lens of mode connectivity, the
observation that minimizers of neural networks retrieved via training on a dataset are …

Unleashing the power of data tsunami: A comprehensive survey on data assessment and selection for instruction tuning of language models

Y Qin, Y Yang, P Guo, G Li, H Shao, Y Shi, Z Xu… - ar**_Avoiding_Confidently_Learning_from_Mislabeled_Examples_ICCV_2023_paper.pdf" data-clk="hl=ja&sa=T&oi=gga&ct=gga&cd=5&d=3650356778833963602&ei=kT2nZ8HfCpiA6rQPmo3bwA0" data-clk-atid="UmrWYSasqDIJ" target="_blank">[PDF] thecvf.com

Late stop**: Avoiding confidently learning from mislabeled examples

S Yuan, L Feng, T Liu - Proceedings of the IEEE/CVF …, 2023 - openaccess.thecvf.com
Sample selection is a prevalent method in learning with noisy labels, where small-loss data
are typically considered as correctly labeled data. However, this method may not effectively …

Can neural network memorization be localized?

P Maini, MC Mozer, H Sedghi, ZC Lipton… - ar** against label noise without validation data
S Yuan, L Feng, T Liu - The Twelfth International Conference on …, 2024 - openreview.net
Early stop** methods in deep learning face the challenge of balancing the volume of
training and validation data, especially in the presence of label noise. Concretely, sparing …

Memorization through the lens of curvature of loss function around samples

I Garg, D Ravikumar, K Roy - arxiv preprint arxiv:2307.05831, 2023 - arxiv.org
Deep neural networks are over-parameterized and easily overfit the datasets they train on.
In the extreme case, it has been shown that these networks can memorize a training set with …