| Challenge: | Existing attention mechanisms are hard and hard, but they are more accurate when trained. |
| Approach: | They propose to use a beam approximation of the joint distribution between attention and output to train sequence to sequence learning. |
| Outcome: | The proposed method is compared to existing attention mechanisms on five translation tasks and shows consistent gains on the same tasks. |
Similar Papers
Classical Structured Prediction Losses for Sequence to Sequence Learning (N18-1)
Copied to clipboard
| Challenge: | Recent work on training neural attention models at the sequence level has focused on a series of objective functions commonly used for structured prediction. |
| Approach: | They propose to use objective functions commonly used to train linear models for structured prediction to train neural attention models at the sequence-level using either reinforcement learning-style methods or beam search optimization. |
| Outcome: | The proposed model outperforms beam search optimization on German-English translation and abstractive summarization tasks. |
Hard Non-Monotonic Attention for Character-Level Transduction (D18-1)
Copied to clipboard
| Challenge: | Character-level string-to-string transductions are an important component of NLP tasks . hard non-monotonic attention models have been used for sequence modeling tasks involving characters . |
| Approach: | They propose an exact algorithm for marginalizing over the exponential number of non-monotonic alignments between two strings. |
| Outcome: | The proposed algorithm outperforms soft attention and improves performance over stochastic approximation. |
Exact Hard Monotonic Attention for Character-Level Transduction (P19-1)
Copied to clipboard
| Challenge: | Neural sequence-to-sequence models with soft attention outperform monotonic models . current dominant method is the neural sequenceto-Sequency model with soft focus . |
| Approach: | They develop a hard attention sequence-to-sequence model that enforces strict monotonicity and learns alignment jointly. |
| Outcome: | The proposed model achieves state-of-the-art on grapheme-to-phoneme conversion and morphological inflection generation. |
Sparse Sequence-to-Sequence Models (P19-1)
Copied to clipboard
| Challenge: | Sequence-to-sequence models are dense and assigning nonzero probability to implausible outputs. |
| Approach: | They propose a new family of -entmax transformations that includes softmax and sparsemax as particular cases and is sparser for any > 1 . they provide fast algorithms to evaluate these transformations and their gradients, which scale well for large vocabulary sizes. |
| Outcome: | The proposed models are able to produce sparse alignments and assign nonzero probability to short list of plausible outputs, sometimes rendering beam search exact. |
Simulating Hard Attention Using Soft Attention (2026.tacl-1)
Copied to clipboard
| Challenge: | a central element of hard attention is attention, which computes a weighted average of values from all unmasked positions. |
| Approach: | They propose transformers that can simulate hard attention by using temperature scaling and positional embeddings. |
| Outcome: | The proposed transformers can effectively focus all attention on a subset of positions. |
Self-Attention with Relative Position Representations (N18-2)
Copied to clipboard
| Challenge: | Recent approaches to sequence to sequence learning leverage recurrence, convolution, attention or combination of recurrent and convolutional neural networks. |
| Approach: | They propose an approach that extends the self-attention mechanism to consider representations of relative positions, or distances between sequence elements. |
| Outcome: | The proposed approach yields 1.3 BLEU and 0.3 BLUE on translation tasks . it is based on a relation-aware self-attention mechanism that can generalize to arbitrary graph-labeled inputs. |
Rectified Sparse Attention for Efficient Long-Sequence Generation (2026.findings-acl)
Copied to clipboard
Yutao Sun, Tianzhu Ye, Li Dong, Yuqing Xia, Jian Chen, Yizhao Gao, Shijie Cao, Jianyong Wang, Furu Wei
| Challenge: | Recent sparse decoding methods improve efficiency but suffer from KV cache misalignment, resulting in performance degradation. |
| Approach: | They propose a method that combines block-sparse attention with periodic dense rectification to bound error accumulation and preserve alignment with the pretraining distribution. |
| Outcome: | Experiments on math reasoning, language modeling, and retrieval tasks show that ReSA achieves near-lossless generation quality with significantly improved efficiency. |
Attention Mechanism with Energy-Friendly Operations (2022.findings-acl)
Copied to clipboard
| Challenge: | Empirical results show that attention mechanism can be improved from the energy consumption aspects. |
| Approach: | They propose to replace multiplications with either selective operations or additions to reduce energy consumption. |
| Outcome: | The proposed model achieves competitable accuracy while saving 99% and 66% energy during alignment calculation and the whole attention procedure. |
Enhancing Machine Translation with Dependency-Aware Self-Attention (2020.acl-main)
Copied to clipboard
| Challenge: | Currently, most neural machine translation models rely on pairs of parallel sentences, assuming syntactic information is automatically learned by an attention mechanism. |
| Approach: | They propose a parameter-free, dependency-aware self-attention mechanism that integrates syntactic knowledge into a Transformer model and propose 'a parameter free approach' they also propose - a novel mechanism that improves translation quality for long sentences and in low-resource scenarios. |
| Outcome: | The proposed approach improves translation quality on English-German and English-Turkish translation tasks and in low-resource scenarios. |
Attention Fusion: a light yet efficient late fusion mechanism for task adaptation in NLU (2022.findings-naacl)
Copied to clipboard
| Challenge: | a recent study has shown that fine-tuning pre-trained models is parameter-inefficient and expensive. |
| Approach: | They propose a task-attuned token module which integrates pre-trained network representations into a pre-trainer. |
| Outcome: | The proposed model trains only 0.0009% of the parameters and is efficient during computation and scalable during deployment. |