Zobrazeno 1 - 10
of 2 169
pro vyhledávání: '"P, LaChapelle"'
Autor:
Brady, Jack, von Kügelgen, Julius, Lachapelle, Sébastien, Buchholz, Simon, Kipf, Thomas, Brendel, Wieland
Learning disentangled representations of concepts and re-composing them in unseen ways is crucial for generalizing to out-of-domain situations. However, the underlying properties of concepts that enable such disentanglement and compositional generali
Externí odkaz:
http://arxiv.org/abs/2411.07784
We analyze identifiability as a possible explanation for the ubiquity of linear properties across language models, such as the vector difference between the representations of "easy" and "easiest" being parallel to that between "lucky" and "luckiest"
Externí odkaz:
http://arxiv.org/abs/2410.23501
Autor:
Brouillard, Philippe, Lachapelle, Sébastien, Kaltenborn, Julia, Gurwicz, Yaniv, Sridhar, Dhanya, Drouin, Alexandre, Nowack, Peer, Runge, Jakob, Rolnick, David
Scientific research often seeks to understand the causal structure underlying high-level variables in a system. For example, climate scientists study how phenomena, such as El Ni\~no, affect other climate processes at remote locations across the glob
Externí odkaz:
http://arxiv.org/abs/2410.07013
Many causal systems such as biological processes in cells can only be observed indirectly via measurements, such as gene expression. Causal representation learning -- the task of correctly mapping low-level observations to latent causal variables --
Externí odkaz:
http://arxiv.org/abs/2405.20482
Autor:
Xu, Danru, Yao, Dingling, Lachapelle, Sébastien, Taslakian, Perouz, von Kügelgen, Julius, Locatello, Francesco, Magliacane, Sara
Causal representation learning aims at identifying high-level causal variables from perceptual data. Most methods assume that all latent causal variables are captured in the high-dimensional observations. We instead consider a partially observed sett
Externí odkaz:
http://arxiv.org/abs/2403.08335
Autor:
Lachapelle, Sébastien, López, Pau Rodríguez, Sharma, Yash, Everett, Katie, Priol, Rémi Le, Lacoste, Alexandre, Lacoste-Julien, Simon
This work introduces a novel principle for disentanglement we call mechanism sparsity regularization, which applies when the latent factors of interest depend sparsely on observed auxiliary variables and/or past latent factors. We propose a represent
Externí odkaz:
http://arxiv.org/abs/2401.04890
Autor:
Yao, Dingling, Xu, Danru, Lachapelle, Sébastien, Magliacane, Sara, Taslakian, Perouz, Martius, Georg, von Kügelgen, Julius, Locatello, Francesco
We present a unified framework for studying the identifiability of representations learned from simultaneously observed views, such as different data modalities. We allow a partially observed setting in which each view constitutes a nonlinear mixture
Externí odkaz:
http://arxiv.org/abs/2311.04056
Publikováno v:
Atmospheric Chemistry and Physics, Vol 24, Pp 11285-11304 (2024)
Ice pellets can form when supercooled raindrops collide with small ice particles that can be generated through secondary ice production processes. The use of atmospheric models that neglect these collisions can lead to an overestimation of freezing r
Externí odkaz:
https://doaj.org/article/e7a05b4f9e9b4419adb4136ad26442c1
We tackle the problems of latent variables identification and ``out-of-support'' image generation in representation learning. We show that both are possible for a class of decoders that we call additive, which are reminiscent of decoders used for obj
Externí odkaz:
http://arxiv.org/abs/2307.02598
Autor:
Lachapelle, Sébastien, Deleu, Tristan, Mahajan, Divyat, Mitliagkas, Ioannis, Bengio, Yoshua, Lacoste-Julien, Simon, Bertrand, Quentin
Although disentangled representations are often said to be beneficial for downstream tasks, current empirical and theoretical understanding is limited. In this work, we provide evidence that disentangled representations coupled with sparse base-predi
Externí odkaz:
http://arxiv.org/abs/2211.14666