ReZero is All You Need: Fast Convergence at Large Depth

• URL: http://arxiv.org/abs/2003.04887v2
• Date: Thu, 25 Jun 2020 00:09:04 GMT
• Title: ReZero is All You Need: Fast Convergence at Large Depth
• Abstract summary: We show that the simplest architecture change of gating each residual connection using a single zero-d parameter satisfies initial dynamical isometry. Although much simpler than its predecessors, this gate enables training thousands of fully connected layers with fast convergence.
• Score: 14.729829527661966
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Deep networks often suffer from vanishing or exploding gradients due to inefficient signal propagation, leading to long training times or convergence difficulties. Various architecture designs, sophisticated residual-style networks, and initialization schemes have been shown to improve deep signal propagation. Recently, Pennington et al. used free probability theory to show that dynamical isometry plays an integral role in efficient deep learning. We show that the simplest architecture change of gating each residual connection using a single zero-initialized parameter satisfies initial dynamical isometry and outperforms more complex approaches. Although much simpler than its predecessors, this gate enables training thousands of fully connected layers with fast convergence and better test performance for ResNets trained on CIFAR-10. We apply this technique to language modeling and find that we can easily train 120-layer Transformers. When applied to 12 layer Transformers, it converges 56% faster on enwiki8.

Related papers

• Replacement Learning: Training Neural Networks with Fewer Parameters [15.09968642484538]
  Replacement Learning (RepL) is a training-time paradigm that reduces full-depth redundancy by replacing selected blocks.<n>RepL preserves local contextual continuity while avoiding unnecessary full-layer.<n>We show that RepL reduces trainable parameters, GPU memory usage, and training time while matching or surpassing standard end-to-end training.
  arXiv  Detail & Related papers  (2026-05-19T08:34:31Z)
• Alternating Gradient Flow Utility: A Unified Metric for Structural Pruning and Dynamic Routing in Deep Networks [52.153950303594684]
  We propose a decoupled kinetic paradigm inspired by Alternating Gradient Flow (AGF)<n>AGF accurately captures the network's structural "kinetic utility"<n>We design a hybrid routing framework that decouples AGF-guided offline structural search from online execution via zero-cost physical priors.
  arXiv  Detail & Related papers  (2026-03-12T18:19:21Z)
• Rethinking Vision Transformer Depth via Structural Reparameterization [16.12815682992294]
  We propose a branch-based structural reparameterization technique that operates during the training phase.<n>Our approach leverages parallel branches within transformer blocks that can be systematically consolidated into streamlined single-path models.<n>When applied to ViT-Tiny, the framework successfully reduces the original 12-layer architecture to 6, 4, or as few as 3 layers while maintaining classification accuracy on ImageNet-1K.
  arXiv  Detail & Related papers  (2025-11-24T21:28:55Z)
• Network Sparsity Unlocks the Scaling Potential of Deep Reinforcement Learning [57.3885832382455]
  We show that introducing static network sparsity alone can unlock further scaling potential beyond dense counterparts with state-of-the-art architectures.<n>Our analysis reveals that, in contrast to naively scaling up dense DRL networks, such sparse networks achieve both higher parameter efficiency for network expressivity.
  arXiv  Detail & Related papers  (2025-06-20T17:54:24Z)
• Auto-Compressing Networks [59.83547898874152]
  We introduce Auto- Networks (ACNs), an architectural variant where additive long feedforward connections from each layer replace traditional short residual connections.<n>ACNs showcase unique property we coin as "auto-compression", the ability of a network to organically compress information during training.<n>We find that ACNs exhibit enhanced noise robustness compared to residual networks, superior performance in low-data settings, and mitigate catastrophic forgetting.
  arXiv  Detail & Related papers  (2025-06-11T13:26:09Z)
• LayerDropBack: A Universally Applicable Approach for Accelerating Training of Deep Networks [5.00301731167245]
  We introduce LayerDropBack (LDB), a simple yet effective method to accelerate training across a wide range of deep networks. LDB introduces randomness only in the backward pass, maintaining the integrity of the forward pass. Our experiments show significant training time reductions of 16.93% to 23.97%, while preserving or even enhancing model accuracy.
  arXiv  Detail & Related papers  (2024-12-23T22:39:41Z)
• Pushing the Limits of Sparsity: A Bag of Tricks for Extreme Pruning [4.421875265386832]
  Pruning of deep neural networks has been an effective technique for reducing model size while preserving most of the performance of dense networks. Recent sparse learning methods have shown promising performance up to moderate sparsity levels such as 95% and 98%. We propose a collection of techniques that enable the continuous learning of networks without accuracy collapse even at extreme sparsities.
  arXiv  Detail & Related papers  (2024-11-20T18:54:53Z)
• Geometry Aware Meta-Learning Neural Network for Joint Phase and Precoder Optimization in RIS [9.20186865054847]
  We propose a complex-valued, geometry aware meta-learning neural network that maximizes the weighted sum rate in a multi-user multiple input single output system. We use a complex-valued neural network for phase shifts and an Euler inspired update for the precoder network. Our approach outperforms existing neural network-based algorithms, offering higher weighted sum rates, lower power consumption, and significantly faster convergence.
  arXiv  Detail & Related papers  (2024-09-17T15:20:23Z)
• Auto-Train-Once: Controller Network Guided Automatic Network Pruning from Scratch [72.26822499434446]
  Auto-Train-Once (ATO) is an innovative network pruning algorithm designed to automatically reduce the computational and storage costs of DNNs. We provide a comprehensive convergence analysis as well as extensive experiments, and the results show that our approach achieves state-of-the-art performance across various model architectures.
  arXiv  Detail & Related papers  (2024-03-21T02:33:37Z)
• Principled Architecture-aware Scaling of Hyperparameters [69.98414153320894]
  Training a high-quality deep neural network requires choosing suitable hyperparameters, which is a non-trivial and expensive process. In this work, we precisely characterize the dependence of initializations and maximal learning rates on the network architecture. We demonstrate that network rankings can be easily changed by better training networks in benchmarks.
  arXiv  Detail & Related papers  (2024-02-27T11:52:49Z)
• Analyzing and Improving the Training Dynamics of Diffusion Models [36.37845647984578]
  We identify and rectify several causes for uneven and ineffective training in the popular ADM diffusion model architecture. We find that systematic application of this philosophy eliminates the observed drifts and imbalances, resulting in considerably better networks at equal computational complexity.
  arXiv  Detail & Related papers  (2023-12-05T11:55:47Z)
• Learning a Consensus Sub-Network with Polarization Regularization and One Pass Training [3.2214522506924093]
  Pruning schemes create extra overhead either by iterative training and fine-tuning for static pruning or repeated computation of a dynamic pruning graph. We propose a new parameter pruning strategy for learning a lighter-weight sub-network that minimizes the energy cost while maintaining comparable performance to the fully parameterised network on given downstream tasks. Our results on CIFAR-10 and CIFAR-100 suggest that our scheme can remove 50% of connections in deep networks with less than 1% reduction in classification accuracy.
  arXiv  Detail & Related papers  (2023-02-17T09:37:17Z)
• The Underlying Correlated Dynamics in Neural Training [6.385006149689549]
  Training of neural networks is a computationally intensive task. We propose a model based on the correlation of the parameters' dynamics, which dramatically reduces the dimensionality. This representation enhances the understanding of the underlying training dynamics and can pave the way for designing better acceleration techniques.
  arXiv  Detail & Related papers  (2022-12-18T08:34:11Z)
• Training Spiking Neural Networks with Local Tandem Learning [96.32026780517097]
  Spiking neural networks (SNNs) are shown to be more biologically plausible and energy efficient than their predecessors. In this paper, we put forward a generalized learning rule, termed Local Tandem Learning (LTL) We demonstrate rapid network convergence within five training epochs on the CIFAR-10 dataset while having low computational complexity.
  arXiv  Detail & Related papers  (2022-10-10T10:05:00Z)
• Manifold Regularized Dynamic Network Pruning [102.24146031250034]
  This paper proposes a new paradigm that dynamically removes redundant filters by embedding the manifold information of all instances into the space of pruned networks. The effectiveness of the proposed method is verified on several benchmarks, which shows better performance in terms of both accuracy and computational cost.
  arXiv  Detail & Related papers  (2021-03-10T03:59:03Z)
• GradInit: Learning to Initialize Neural Networks for Stable and Efficient Training [59.160154997555956]
  We present GradInit, an automated and architecture method for initializing neural networks. It is based on a simple agnostic; the variance of each network layer is adjusted so that a single step of SGD or Adam results in the smallest possible loss value. It also enables training the original Post-LN Transformer for machine translation without learning rate warmup.
  arXiv  Detail & Related papers  (2021-02-16T11:45:35Z)
• A Convergence Theory Towards Practical Over-parameterized Deep Neural Networks [56.084798078072396]
  We take a step towards closing the gap between theory and practice by significantly improving the known theoretical bounds on both the network width and the convergence time. We show that convergence to a global minimum is guaranteed for networks with quadratic widths in the sample size and linear in their depth at a time logarithmic in both. Our analysis and convergence bounds are derived via the construction of a surrogate network with fixed activation patterns that can be transformed at any time to an equivalent ReLU network of a reasonable size.
  arXiv  Detail & Related papers  (2021-01-12T00:40:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.