paper-with-me

Papers

Deep Learning without Shortcuts: Shaping the Kernel with Tailored Rectifiers

2022-03-15 · ICLR 2022 4 · Guodong Zhang, Aleksandar Botev, James Martens

Training very deep neural networks is still an extremely challenging task. The common solution is to use shortcut connections and normalization layers, which are both crucial ingredients in the popular ResNet architecture. However, there is strong evidence to suggest that ResNets behave more like ensembles of shallower networks than truly deep ones. Recently, it was shown that deep vanilla networks (i.e. networks without normalization layers or shortcut connections) can be trained as fast as ResNets by applying certain transformations to their activation functions. However, this method (called Deep Kernel Shaping) isn't fully compatible with ReLUs, and produces networks that overfit significantly more than ResNets on ImageNet. In this work, we rectify this situation by developing a new type of transformation that is fully compatible with a variant of ReLUs -- Leaky ReLUs. We show in experiments that our method, which introduces negligible extra computational cost, achieves validation accuracies with deep vanilla networks that are competitive with ResNets (of the same width/depth), and significantly higher than those obtained with the Edge of Chaos (EOC) method. And unlike with EOC, the validation accuracies we obtain do not get worse with depth.

📄 PDF Abstract BibTeX arXiv:2203.08120

Code (1)

deepmind/dks 공식 구현 jax

Tasks

Deep Learning

Methods 이 논문이 사용한 방법론

ReLU How Do I Communicate to Expedia? How Do I Communicate to Expedia? – Call ☎️ +1-(888) 829 (0881) or +1-805-330-4056 or +1-805-330-4056 for Live Support & Special Travel…
Batch Normalization 설명 없음
Residual Connection 설명 없음
1x1 Convolution A 1 x 1 Convolution is a convolution with some special properties in that it can be used for dimensionality reduction,…
Average Pooling 설명 없음
Bottleneck Residual Block A Bottleneck Residual Block is a variant of the residual block that utilises 1x1 convolutions to create a bottleneck. The…
Kaiming Initialization 설명 없음
Max Pooling Max Pooling is a pooling operation that calculates the maximum value for patches of a feature map, and uses it to create a downsampled (pooled) feature map. It is usually…

Similar Papers 제목 키워드 기반

Deep Transformers without Shortcuts: Modifying Self-attention for Faithful Signal Propagation

2023-02-20 · Bobby He, James Martens, Guodong Zhang, Aleksandar Botev 외

Skip connections and normalisation layers form two standard architectural components that are ubiquitous for the training of Deep Neural Networks (DNNs), but whose precise roles are poorly understood. Recent approaches s…

Optimal Switching of Controlled Rectifiers

2020-02-25

This paper discusses a linear programming approach for designing switching signals for controlled rectifiers to achieve a low input current & output voltage total harmonic distortions. The focus here is on fully controll…

Rapid training of deep neural networks without skip connections or normalization layers using Deep Kernel Shaping

2021-10-05 · James Martens, Andy Ballard, Guillaume Desjardins, Grzegorz Swirszcz 외

Using an extended and formalized version of the Q/C map analysis of Poole et al. (2016), along with Neural Tangent Kernel theory, we identify the main pathologies present in deep networks that prevent them from training …

Navigating Shortcuts, Spurious Correlations, and Confounders: From Origins via Detection to Mitigation

2024-12-06 · David Steinmann, Felix Divo, Maurice Kraus, Antonia Wüst 외

Shortcuts, also described as Clever Hans behavior, spurious correlations, or confounders, present a significant challenge in machine learning and AI, critically affecting model generalization and robustness. Research in …

Overcoming Shortcut Learning in Graph Neural Networks through Active Explanation Guidance

2026-08-14 · Taraneh Younesian, Steve Azzolin, Antonio Longa, Francesco Ferrini 외 arxiv

Graph Neural Networks (GNNs) can solve prediction tasks by unintentionally exploiting shortcuts---that is, edges, nodes, and features that correlate with but are not causal for the prediction---which compromise their rel…

Active Learning