Publications related to On Symmetry and Initialization for Neural Networks

Related publications (14)

Investigating OCR-Sensitive Neurons to Improve Entity Recognition in Historical Documents

This paper investigates the presence of OCR-sensitive neurons within the Transformer architecture and their influence on named entity recognition (NER) performance on historical documents. By analysing neuron activation patterns in response to clean and no ...

Springer Science and Business Media Deutschland GmbH2025

On the impact of learning-based image compression on computer vision tasks

Touradj Ebrahimi, Evgeniy Upenik, Michela Testolina

The image compression field is witnessing a shift in paradigm thanks to the rise of neural network-based models. In this context, the JPEG committee is in the process of standardizing the first learning-based image compression standard, known as JPEG AI. W ...

2024

Automatic Parameterization for Aerodynamic Shape Optimization via Deep Geometric Learning

Pascal Fua, Zhen Wei

We propose two deep learning models that fully automate shape parameterization for aerodynamic shape optimization. Both models are optimized to parameterize via deep geometric learning to embed human prior knowledge into learned geometric patterns, elimina ...

2023

Estimating cross-field particle transport at the outer midplane of TCV by tracking filaments with machine learning

Christian Gabriel Theiler, Curdin Tobias Wüthrich, Nicola Offeddu, Davide Galassi, Claudia Colandrea

Cross-field transport of particles in the boundary region of magnetically confined fusion plasmas is dominated by turbulence. Blobs, intermittent turbulent structures with large amplitude and a filamentary shape appearing in the scrape-off layer (SOL), are ...

IOP Publishing Ltd2023

Thermalization and chaos in a 1+1d QFT

Matthew Thomas Walters

We study aspects of chaos and thermodynamics at strong coupling in a scalar model using LCT numerical methods. We find that our eigenstate spectrum satisfies Wigner-Dyson statistics and that the coefficients describing eigenstates in our basis satisfy Rand ...

Springer Nature2023

Brain signals of a Surprise-Actor-Critic model: Evidence for multiple learning modules in human decision making

Wulfram Gerstner, Marco Philipp Lehmann, Vasiliki Liakoni, Kerstin Preuschoff, Johanni Michael Brea, Alireza Modirshanechi

Learning how to reach a reward over long series of actions is a remarkable capability of humans, and potentially guided by multiple parallel learning modules. Current brain imaging of learning modules is limited by (i) simple experimental paradigms, (ii) e ...

ACADEMIC PRESS INC ELSEVIER SCIENCE2022

Perceiving Humans: from Monocular 3D Localization to Social Distancing

Alexandre Massoud Alahi, Sven Kreiss, Lorenzo Bertoni

Perceiving humans in the context of Intelligent Transportation Systems (ITS) often relies on multiple cameras or expensive LiDAR sensors. In this work, we present a new cost-effective vision-based method that perceives humans' locations in 3D and their bod ...

IEEE Institute of Electrical and Electronics Engineers2022

Deep transfer learning of additive manufacturing mechanisms across materials in metal-based laser powder bed fusion process

Rita Drissi Daoudi, Roland Logé

The defective regimes in metal-based Laser Powder Bed Fusion (LPBF) processes can be minimized by deploying in-situ monitoring strategies comprising Machine learning (ML) algorithms and sensing techniques. So far, algorithms trained for monitoring a partic ...

ELSEVIER SCIENCE SA2022

Robustness via Cross-Domain Ensembles

Amir Roshan Zamir, Shuqing Teresa Yeo, Oguzhan Fatih Kar

We present a method for making neural network predictions robust to shifts from the training data distribution. The proposed method is based on making predictions via a diverse set of cues (called 'middle domains') and ensembling them into one strong predi ...

IEEE2021

Training Channel Selection for Learning-based 1-bit Precoding in Massive MU-MIMO

Andreas Peter Burg, Alexios Konstantinos Balatsoukas Stimming, Sitian Li

Learning-based algorithms have gained great popularity in communications since they often outperform even carefully engineered solutions by learning from training samples. In this paper, we show that the selection of appropriate training examples can be im ...

IEEE2020

Mean-Field Games And Swarms Dynamics In Gaussian And Non-Gaussian Environments

Max-Olivier Hongler

The collective behaviour of stochastic multi-agents swarms driven by Gaussian and non-Gaussian environments is analytically discussed in a mean-field approach. We first exogenously implement long range mutual interactions rules with strengths that are weig ...

AMER INST MATHEMATICAL SCIENCES-AIMS2020

Detecting the Unexpected via Image Resynthesis

Pascal Fua, Mathieu Salzmann, Krzysztof Maciej Lis, Krishna Kanth Nakka

Classical semantic segmentation methods, including the recent deep learning ones, assume that all classes observed at test time have been seen during training. In this paper, we tackle the more realistic scenario where unexpected objects of unknown classes ...

IEEE2019

Dilaton-assisted generation of the Fermi scale from the Planck scale

Andrey Shkerin

In scale-invariant theories of gravity the Planck mass M-p, which appears due to spontaneous symmetry breaking, can be the only scale at the classical level. It was argued that the second scale can be generated by a quantum nonperturbative gravitational ef ...

American Physical Society2019

Full-Pulse Tomographic Reconstruction with Deep Neural Networks

Patrick Blanchard, Basil Duval, Ambrogio Fasoli, Yves Martin, Richard Pitts, Olivier Sauter, Henri Weisen, Robin Humphry-Baker, Stefano Coda, Martinus Adela Maria Gijs, Partha Dutta, Marco Wischmeier, Jan Horacek, Vlad Trifa, Yann Camenen, Jonathan Graves, Duccio Testa, Miguel Fernández Ruiz, Sun Hee Kim, Mikhail Maslov, Javier García Hernández, David Pfefferlé, José Pedro Rebelo Ferreira Marques, Julio Rodriguez, Pedro Camilo de Oliveira e Silva, Rebecca Hill, Daniel Scott Alessi, Jonnathan Cesar Hidalgo Acosta, Ana Francisca Leal Silva Soares, Dalziel Joseph Wilson, Pierre-Thomas Paul Brun, Jonathan Marc Philippe Faustin, Federico Nespoli, Alberto Hernando de Castro, Hamish William Patten, Bruno Emanuel Ferreira De Sousa Correia, Li Shuai, Nicola Vianello, Antonio José Pereira de Figueiredo, Samuel Lanthaler, Liang Yao, Arnout Lodewijk M Beckers, Cristian Sommariva, Davide Galassi, Alessandro Pau

Plasma tomography consists of reconstructing a two-dimensional radiation profile of a poloidal cross section of a fusion device based on line-integrated measurements along several lines of sight. The reconstruction process is computationally intensive, and ...

2018