Documentdetail
ID kaart
oai:arXiv.org:2410.16868
Onderwerp
Computer Science - Machine Learnin... Computer Science - Computer Vision...Auteur
Martinetz, Julius Linse, Christoph Martinetz, ThomasCategorie
Computer Science
Jaar
2024
vermelding datum
30-10-2024
Trefwoorden
separable scenarios classes computer training classifiersBeschrijving
We investigate the learning dynamics of classifiers in scenarios where classes are separable or classifiers are over-parameterized.
In both cases, Empirical Risk Minimization (ERM) results in zero training error.
However, there are many global minima with a training error of zero, some of which generalize well and some of which do not.
We show that in separable classes scenarios the proportion of "bad" global minima diminishes exponentially with the number of training data n.
Our analysis provides bounds and learning curves dependent solely on the density distribution of the true error for the given classifier function set, irrespective of the set's size or complexity (e.g., number of parameters).
This observation may shed light on the unexpectedly good generalization of over-parameterized Neural Networks.
For the over-parameterized scenario, we propose a model for the density distribution of the true error, yielding learning curves that align with experiments on MNIST and CIFAR-10.
Martinetz, Julius,Linse, Christoph,Martinetz, Thomas, 2024, Rethinking generalization of classifiers in separable classes scenarios and over-parameterized regimes
Multi-agent cooperation through learning-aware policy gradients
self-interested learning-aware cooperation learning
Prodromal Parkinson disease signs are predicted by a whole-blood inflammatory transcriptional signature in young Pink1(−/−) rats
−/− data rats expression pink1 disease blood gene
