Zobrazeno 1 - 10
of 732
pro vyhledávání: '"Baratin A"'
Generating novel molecules is challenging, with most representations leading to generative models producing many invalid molecules. Spanning Tree-based Graph Generation (STGG) is a promising approach to ensure the generation of valid molecules, outpe
Externí odkaz:
http://arxiv.org/abs/2407.09357
Machine learning systems often acquire biases by leveraging undesired features in the data, impacting accuracy variably across different sub-populations. Current understanding of bias formation mostly focuses on the initial and final stages of learni
Externí odkaz:
http://arxiv.org/abs/2405.18296
Autor:
Dufort-Labbé, Simon, D'Oro, Pierluca, Nikishin, Evgenii, Pascanu, Razvan, Bacon, Pierre-Luc, Baratin, Aristide
When training deep neural networks, the phenomenon of $\textit{dying neurons}$ $\unicode{x2013}$units that become inactive or saturated, output zero during training$\unicode{x2013}$ has traditionally been viewed as undesirable, linked with optimizati
Externí odkaz:
http://arxiv.org/abs/2403.07688
Autor:
Arefin, Md Rifat, Zhang, Yan, Baratin, Aristide, Locatello, Francesco, Rish, Irina, Liu, Dianbo, Kawaguchi, Kenji
Publikováno v:
ICLM 2024
Models prone to spurious correlations in training data often produce brittle predictions and introduce unintended biases. Addressing this challenge typically involves methods relying on prior knowledge and group annotation to remove spurious correlat
Externí odkaz:
http://arxiv.org/abs/2402.13368
Autor:
Liu, Yuhan Helena, Baratin, Aristide, Cornford, Jonathan, Mihalas, Stefan, Shea-Brown, Eric, Lajoie, Guillaume
In theoretical neuroscience, recent work leverages deep learning tools to explore how some network attributes critically influence its learning dynamics. Notably, initial weight distributions with small (resp. large) variance may yield a rich (resp.
Externí odkaz:
http://arxiv.org/abs/2310.08513
Sharpness-aware minimization (SAM) methods have gained increasing popularity by formulating the problem of minimizing both loss value and loss sharpness as a minimax objective. In this work, we increase the efficiency of the maximization and minimiza
Externí odkaz:
http://arxiv.org/abs/2307.16704
Autor:
Malviya, Pranshu, Mordido, Gonçalo, Baratin, Aristide, Harikandeh, Reza Babanezhad, Huang, Jerry, Lacoste-Julien, Simon, Pascanu, Razvan, Chandar, Sarath
Adaptive gradient-based optimizers, notably Adam, have left their mark in training large-scale deep learning models, offering fast convergence and robustness to hyperparameter settings. However, they often struggle with generalization, attributed to
Externí odkaz:
http://arxiv.org/abs/2307.09638
We approach the problem of improving robustness of deep learning algorithms in the presence of label noise. Building upon existing label correction and co-teaching methods, we propose a novel training procedure to mitigate the memorization of noisy l
Externí odkaz:
http://arxiv.org/abs/2212.01674
Publikováno v:
Studies in Digital Heritage, Vol 8, Iss 1 (2024)
This article describes the methodology used to automatically classify the constituent elements of a monochromatic mosaic from the Roman era. The classification involved the recognition of the tesserae (fragments of black and white materials for the f
Externí odkaz:
https://doaj.org/article/150fb033964041b3ad3bffe925007123
Publikováno v:
TMLR 2022 - Transactions on Machine Learning Research, 12/2022
Among attempts at giving a theoretical account of the success of deep neural networks, a recent line of work has identified a so-called lazy training regime in which the network can be well approximated by its linearization around initialization. Her
Externí odkaz:
http://arxiv.org/abs/2209.09658