Challenge: Existing studies show that deep Transformers have difficulty in training even with residual connection and layer normalization.
Approach: They propose a method that leverages the Lipschitz constraint on the initialization of Transformer parameters to ease the optimization difficulties caused by its multi-layer encoder/decoder structure.
Outcome: The proposed model outperforms previous RNN/CNN models but fails to converge with the original computation order.

Similar Papers

Improving Deep Transformer with Depth-Scaled Initialization and Merged Attention (D19-1)

Copied to clipboard

Challenge: Existing methods to improve NLP convergence and computational overhead are limited by stacking more layers.
Approach: They propose a depth-scaled initialization method which reduces parameter variance at initialization and reduces output variance of residual connections to ease gradient back-propagation.
Outcome: The proposed method outperforms the base model on translation tasks with five translation directions while matching the decoding speed of the baseline model.
Learning Deep Transformer Models for Machine Translation (P19-1)

Copied to clipboard

Challenge: Neural machine translation models have advanced the previous state-of-the-art by learning mappings between sequences via neural networks and attention mechanisms.
Approach: They propose to use layer normalization to pass the combination of previous layers to the next layer to improve the model.
Outcome: The proposed model outperforms the shallow Transformer-Big/Base baseline model on English-German and Chinese-English tasks by 0.4-2.4 BLEU points.
B2T Connection: Serving Stability and Performance in Deep Transformers (2023.findings-acl)

Copied to clipboard

Challenge: Existing methods to prevent the vanishing gradient problem in deep neural networks are not effective.
Approach: They propose a method that can equip both higher stability and effective training by a simple modification from Post-LN.
Outcome: The proposed method outperforms Pre-LN and Post-Ln on a wide range of tasks.
Optimizing Deeper Transformers on Small Datasets (2021.acl-long)

Copied to clipboard

Challenge: a common belief that training deep transformers from scratch requires large datasets is wrong . however, with proper initialization and optimization, the benefits of very deep transformer can carry over to challenging tasks with small datasets.
Approach: They train 48 layers of transformers from pre-trained RoBERTa and 24 relation-aware layers from scratch.
Outcome: The proposed scheme achieves state-of-the-art performance on a text-to-sql parsing benchmark . it uses 24 fine-tuned layers from pre-trained RoBERTa and 24 relation-aware layers from scratch .
Rewiring the Transformer with Depth-Wise LSTMs (2024.lrec-main)

Copied to clipboard

Challenge: Stacking non-linear layers allows deep neural networks to model complicated functions . but residual connections within each layer fail to fuse information from previous layers effectively .
Approach: They propose a Transformer with depth-wise LSTMs connecting cascading Transformer layers and sub-layers.
Outcome: The proposed model improves in English-German / French and multilingual tasks with BLEU.
Understanding the Difficulty of Training Transformers (2020.emnlp-main)

Copied to clipboard

Challenge: Admin (Adaptive model initialization) is more stable, converges faster, and leads to better performance.
Approach: They propose a model initialization algorithm to stabilize early training and unleash its full potential in the late stage.
Outcome: The proposed model initialization method stabilizes early training and unleashes full potential in late stage.
Exploring Quantization for Efficient Pre-Training of Transformer Language Models (2024.findings-emnlp)

Copied to clipboard

Challenge: Quantization has proven to be effective after pre-training and during fine-tuning, but its effects on pre-trainer performance have remained unexplored.
Approach: They propose a linear quantization strategy to be applied during the pre-training of Transformers to improve model efficiency and stability.
Outcome: The proposed method improves model efficiency, stability, and performance while maintaining language modeling ability.
Sneaking Syntax into Transformer Language Models with Tree Regularization (2025.naacl-long)

Copied to clipboard

Challenge: Existing methods for incorporating syntactic inductive biases into transformers are limited . we introduce auxiliary loss function that converts bracketing decisions into differentiable orthogonality constraints on vector hidden states.
Approach: They propose to introduce syntactic inductive biases into transformer circuits through a structured regularizer.
Outcome: The proposed approach could unlock more robust and data-efficient learning in transformer language models . it integrates seamlessly with the standard LM objective, requiring no architectural changes.
Incorporating Residual and Normalization Layers into Analysis of Masked Language Models (2021.emnlp-main)

Copied to clipboard

Challenge: Transformer architecture is composed of multi-head attention, which has been extensively analyzed.
Approach: They extended the scope of the analysis of Transformers from solely the attention patterns to the whole attention block, i.e., multi-head attention, residual connection, and layer normalization.
Outcome: The proposed method incorporates the whole attention block, i.e., multi-head attention, residual connection, and layer normalization into the analysis.
Subformer: Exploring Weight Sharing for Parameter Efficiency in Generative Transformers (2021.findings-emnlp)

Copied to clipboard

Challenge: Recent improvements in NLP tasks can be attributed to the Transformer model.
Approach: They propose to use parameter-sharing methods to reduce parameter budgets in generative models by using sandwich-style parameter sharing and self-attentive embedding factorization.
Outcome: The proposed model outperforms the current RNN model even with significantly fewer parameters.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations