Zobrazeno 1 - 10
of 337
pro vyhledávání: '"Refinetti P"'
Autor:
Srinivasan, Ravi, Mignacco, Francesca, Sorbaro, Martino, Refinetti, Maria, Cooper, Avi, Kreiman, Gabriel, Dellaferrera, Giorgia
"Forward-only" algorithms, which train neural networks while avoiding a backward pass, have recently gained attention as a way of solving the biologically unrealistic aspects of backpropagation. Here, we first address compelling challenges related to
Externí odkaz:
http://arxiv.org/abs/2302.05440
Publikováno v:
ICML 2023
The ability of deep neural networks to generalise well even when they interpolate their training data has been explained using various "simplicity biases". These theories postulate that neural networks avoid overfitting by first learning simple funct
Externí odkaz:
http://arxiv.org/abs/2211.11567
Learning rate schedules are ubiquitously used to speed up and improve optimisation. Many different policies have been introduced on an empirical basis, and theoretical analyses have been developed for convex settings. However, in many realistic probl
Externí odkaz:
http://arxiv.org/abs/2202.04509
Publikováno v:
Proceedings of the 39th International Conference on Machine Learning (ICML). PMLR 162:14283-14314, 2022
From the sampling of data to the initialisation of parameters, randomness is ubiquitous in modern Machine Learning practice. Understanding the statistical fluctuations engendered by the different sources of randomness in prediction is therefore key t
Externí odkaz:
http://arxiv.org/abs/2201.13383
Autor:
Refinetti, Maria, Goldt, Sebastian
Publikováno v:
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:18499-18519 (2022)
Autoencoders are the simplest neural network for unsupervised learning, and thus an ideal framework for studying feature learning. While a detailed understanding of the dynamics of linear autoencoders has recently been obtained, the study of non-line
Externí odkaz:
http://arxiv.org/abs/2201.02115
Publikováno v:
Proceedings of the 38th International Conference on Machine Learning, PMLR 139, 2021
A recent series of theoretical works showed that the dynamics of neural networks with a certain initialisation are well-captured by kernel methods. Concurrent empirical work demonstrated that kernel methods can come close to the performance of neural
Externí odkaz:
http://arxiv.org/abs/2102.11742
Publikováno v:
SciPost Phys. 14, 068 (2023)
We use numerical bootstrap techniques to study correlation functions of traceless symmetric tensors of $O(N)$ with two indexes $t_{ij}$. We obtain upper bounds on operator dimensions for all the relevant representations and several values of $N$. We
Externí odkaz:
http://arxiv.org/abs/2012.08533
Publikováno v:
Proceedings of the 38th International Conference on Machine Learning (ICML), PMLR 139, 2021
Direct Feedback Alignment (DFA) is emerging as an efficient and biologically plausible alternative to the ubiquitous backpropagation algorithm for training deep neural networks. Despite relying on random feedback weights for the backward pass, DFA su
Externí odkaz:
http://arxiv.org/abs/2011.12428
Autor:
Baker, Antoine, Biazzo, Indaco, Braunstein, Alfredo, Catania, Giovanni, Dall'Asta, Luca, Ingrosso, Alessandro, Krzakala, Florent, Mazza, Fabio, Mézard, Marc, Muntoni, Anna Paola, Refinetti, Maria, Mannelli, Stefano Sarao, Zdeborová, Lenka
Publikováno v:
PNAS 2021 Vol. 118 No. 32 e2106548118
Contact-tracing is an essential tool in order to mitigate the impact of pandemic such as the COVID-19. In order to achieve efficient and scalable contact-tracing in real time, digital devices can play an important role. While a lot of attention has b
Externí odkaz:
http://arxiv.org/abs/2009.09422
Deep neural networks can achieve remarkable generalization performances while interpolating the training data perfectly. Rather than the U-curve emblematic of the bias-variance trade-off, their test error often follows a "double descent" - a mark of
Externí odkaz:
http://arxiv.org/abs/2003.01054