LSTMs and other RNN variants have shown strong performance on character-level language modeling. These models are typically trained using truncated backpropagation through time, and it is common to assume that their success stems from their ability to remember long-term contexts. In this paper, we show that a deep (64-layer) transformer model with fixed context outperforms RNN variants by a large margin, achieving state of the art on two popular benchmarks: 1.13 bits per character on text8 and 1.06 on enwik8. To get good results at this depth, we show that it is important to add auxiliary losses, both at intermediate network layers and intermediate sequence positions.
While current state-of-the-art NMT models, such as RNN seq2seq and Transformers, possess a large number of parameters, they are still shallow in comparison to convolutional models used for both text and vision applications. In this work we attempt to train significantly (2-3x) deeper Transformer and Bi-RNN encoders for machine translation. We propose a simple modification to the attention mechanism that eases the optimization of deeper models, and results in consistent gains of 0.7-1.1 BLEU on the benchmark WMT'14 English-German and WMT'15 Czech-English tasks for both architectures.
Uppsala University（乌普萨拉大学）；University of Zurich（苏黎世大学）；University of Edinburgh（爱丁堡大学）
Recent work has shown that deeper character-based neural machine translation (NMT) models can outperform subword-based models. However, it is still unclear what makes deeper character-based models successful. In this paper, we conduct an investigation into pure character-based models in the case of translating Finnish into English, including exploring the ability to learn word senses and morphological inflections and the attention mechanism. We demonstrate that word-level information is distributed over the entire character sequence rather than over a single character, and characters at differ…
The Hierarchical Attention Network (HAN) has made great strides, but it suffers a major limitation: at level 1, each sentence is encoded in complete isolation. In this work, we propose and compare several modifications of HAN in which the sentence encoder is able to make context-aware attentional decisions (CAHAN). Furthermore, we propose a bidirectional document encoder that processes the document forwards and backwards, using the preceding and following sentences as context. Experiments on three large-scale sentiment and topic classification datasets show that the bidirectional version of CA…
We focus on the recognition of Dyck-n ($\mathcal{D}_n$) languages with self-attention (SA) networks, which has been deemed to be a difficult task for these networks. We compare the performance of two variants of SA, one with a starting symbol (SA$^+$) and one without (SA$^-$). Our results show that SA$^+$ is able to generalize to longer sequences and deeper dependencies. For $\mathcal{D}_2$, we find that SA$^-$ completely breaks down on long sequences whereas the accuracy of SA$^+$ is 58.82$\%$. We find attention maps learned by $\text{SA}{^+}$ to be amenable to interpretation and compatible w…
Aspect-level sentiment classification aims to identify the sentiment polarity towards a specific aspect term in a sentence. Most current approaches mainly consider the semantic information by utilizing attention mechanisms to capture the interactions between the context and the aspect term. In this paper, we propose to employ graph convolutional networks (GCNs) on the dependency tree to learn syntax-aware representations of aspect terms. GCNs often show the best performance with two layers, and deeper GCNs do not bring additional gain due to over-smoothing problem. However, in some cases, impo…
Recently multimodal transformer models have gained popularity because their performance on language and vision tasks suggest they learn rich visual-linguistic representations. Focusing on zero-shot image retrieval tasks, we study three important factors which can impact the quality of learned representations: pretraining data, the attention mechanism, and loss functions. By pretraining models on six datasets, we observe that dataset noise and language similarity to our downstream task are important indicators of model performance. Through architectural analysis, we learn that models with a mul…
Empirical Analysis of Decoding Biases in Masked Diffusion Models
School of Computer Science and Engineering, Northeastern University, China（东北大学计算机科学与工程学院）；Department of Computer Science and Technology, Institute for AI, Tsinghua University, China（清华大学人工智能研究院计算机科学与技术系）；Alibaba Group, Hangzhou, China（阿里巴巴集团）
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Google Research（谷歌研究院）
Multi-head attention plays a crucial role in the recent success of Transformer models, which leads to consistent performance improvements over conventional attention in various applications. The popular belief is that this effectiveness stems from the ability of jointly attending multiple positions. In this paper, we first demonstrate that jointly attending multiple positions is not a unique feature of multi-head attention, as multi-layer single-head attention also attends multiple positions and is more effective. Then, we suggest the main advantage of the multi-head attention is the training …
State-of-the-art speaker verification frameworks have typically focused on developing models with increasingly deeper (more layers) and wider (number of channels) models to improve their verification performance. Instead, this paper proposes an approach to increase the model resolution capability using attention-based dynamic kernels in a convolutional neural network to adapt the model parameters to be feature-conditioned. The attention weights on the kernels are further distilled by channel attention and multi-layer feature aggregation to learn global features from speech. This approach provi…
ADAPT Centre（ADAPT中心）；Dublin City University（都柏林城市大学）；Soochow University（苏州大学）
In neural machine translation (NMT), researchers face the challenge of un-seen (or out-of-vocabulary OOV) words translation. To solve this, some researchers propose the splitting of western languages such as English and German into sub-words or compounds. In this paper, we try to address this OOV issue and improve the NMT adequacy with a harder language Chinese whose characters are even more sophisticated in composition. We integrate the Chinese radicals into the NMT model with different settings to address the unseen words challenge in Chinese to English translation. On the other hand, this a…
Multi-Step Deductive Reasoning Over Natural Language: An Empirical Study on Out-of-Distribution Generalisation
Strong AI Lab, NAOInstitute, Waipapa Taumata Rau - The University of Auckland（强人工智能实验室、NAO研究所、怀帕帕·陶马陶·劳 - 奥克兰大学）；Xtracta（埃克斯特拉克塔公司）；Department of Computer Science, University of Cambridge（剑桥大学计算机科学系）
University of Edinburgh（爱丁堡大学）；University of Amsterdam（阿姆斯特丹大学）；University of Zurich（苏黎世大学）
The general trend in NLP is towards increasing model capacity and performance via deeper neural networks. However, simply stacking more layers of the popular Transformer architecture for machine translation results in poor convergence and high computational overhead. Our empirical analysis suggests that convergence is poor due to gradient vanishing caused by the interaction between residual connections and layer normalization. We propose depth-scaled initialization (DS-Init), which decreases parameter variance at the initialization stage, and reduces output variance of residual connections so …
In this paper, we describe the use of recurrent neural networks to capture sequential information from the self-attention representations to improve the Transformers. Although self-attention mechanism provides a means to exploit long context, the sequential information, i.e. the arrangement of tokens, is not explicitly captured. We propose to cascade the recurrent neural networks to the Transformers, which referred to as the TransfoRNN model, to capture the sequential information. We found that the TransfoRNN models which consists of only shallow Transformers stack is suffice to give comparabl…
Garbage Attention in Large Language Models: BOS Sink Heads and Sink-aware Pruning
Graduate School of Data Science, Seoul National University（数据科学研究生院，首尔国立大学）；Department of Rural Systems Engineering, Seoul National University（农村系统工程系，首尔国立大学）；Department of Aerospace Engineering, Seoul National University（航空航天工程系，首尔国立大学）
Transformer has shown promising results in many sequence to sequence transformation tasks recently. It utilizes a number of feed-forward self-attention layers to replace the recurrent neural networks (RNN) in attention-based encoder decoder (AED) architecture. Self-attention layer learns temporal dependence by incorporating sinusoidal positional embedding of tokens in a sequence for parallel computing. Quicker iteration speed in training than sequential operation of RNN can be obtained. Deeper layers of the transformer also make it perform better than RNN-based AED. However, this parallelizati…
