Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userEL-Attention: Memory Efficient Lossless Attention for Generation
102
0
0.0
(
0
)
Added by
Yu Yan
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Transformer model with multi-head attention requires caching intermediate results for efficient inference in generation tasks. However, cache brings new memory-related costs and prevents leveraging larger batch size for faster speed. We propose memory-efficient lossless attention (called EL-attention) to address this issue. It avoids heavy operations for building multi-head keys and values, cache for them is not needed. EL-attention constructs an ensemble of attention results by expanding query while keeping key and value shared. It produces the same result as multi-head attention with less GPU memory and faster inference speed. We conduct extensive experiments on Transformer, BART, and GPT-2 for summarization and question generation tasks. The results show EL-attention speeds up existing models by 1.6x to 5.3x without accuracy loss.
rate research
Read More
Following the success of dot-product attention in Transformers, numerous approximations have been recently proposed to address its quadratic complexity with respect to the input length. While these variants are memory and compute efficient, it is not possible to directly use them with popular pre-trained language models trained using vanilla attention, without an expensive corrective pre-training stage. In this work, we propose a simple yet highly accurate approximation for vanilla attention. We process the queries in chunks, and for each query, compute the top-$k$ scores with respect to the keys. Our approach offers several advantages: (a) its memory usage is linear in the input size, similar to linear attention variants, such as Performer and RFA (b) it is a drop-in replacement for vanilla attention that does not require any corrective pre-training, and (c) it can also lead to significant memory savings in the feed-forward layers after casting them into the familiar query-key-value framework. We evaluate the quality of top-$k$ approximation for multi-head attention layers on the Long Range Arena Benchmark, and for feed-forward layers of T5 and UnifiedQA on multiple QA datasets. We show our approach leads to accuracy that is nearly-identical to vanilla attention in multiple setups including training from scratch, fine-tuning, and zero-shot inference.
Recent advancements in deep learning have created many opportunities to solve real-world problems that remained unsolved for more than a decade. Automatic caption generation is a major research field, and the research community has done a lot of work on it in most common languages like English. Urdu is the national language of Pakistan and also much spoken and understood in the sub-continent region of Pakistan-India, and yet no work has been done for Urdu language caption generation. Our research aims to fill this gap by developing an attention-based deep learning model using techniques of sequence modeling specialized for the Urdu language. We have prepared a dataset in the Urdu language by translating a subset of the Flickr8k dataset containing 700 man images. We evaluate our proposed technique on this dataset and show that it can achieve a BLEU score of 0.83 in the Urdu language. We improve on the previous state-of-the-art by using better CNN architectures and optimization techniques. Furthermore, we provide a discussion on how the generated captions can be made correct grammar-wise.
In this work, we propose three explainable deep learning architectures to automatically detect patients with Alzheimer`s disease based on their language abilities. The architectures use: (1) only the part-of-speech features; (2) only language embedding features and (3) both of these feature classes via a unified architecture. We use self-attention mechanisms and interpretable 1-dimensional ConvolutionalNeural Network (CNN) to generate two types of explanations of the model`s action: intra-class explanation and inter-class explanation. The inter-class explanation captures the relative importance of each of the different features in that class, while the inter-class explanation captures the relative importance between the classes. Note that although we have considered two classes of features in this paper, the architecture is easily expandable to more classes because of its modularity. Extensive experimentation and comparison with several recent models show that our method outperforms these methods with an accuracy of 92.2% and F1 score of 0.952on the DementiaBank dataset while being able to generate explanations. We show by examples, how to generate these explanations using attention values.
Despite considerable advancements with deep neural language models (LMs), neural text generation still suffers from degeneration: generated text is repetitive, generic, self-inconsistent, and lacking commonsense. The empirical analyses on sentence-level attention patterns reveal that neural text degeneration may be associated with insufficient learning of inductive biases by the attention mechanism. Our findings motivate on-the-fly attention modularization, a simple but effective method for injecting inductive biases into attention computation during inference. The resulting text produced by the language model with attention modularization can yield enhanced diversity and commonsense reasoning while maintaining fluency and coherence.
Document grounded generation is the task of using the information provided in a document to improve text generation. This work focuses on two different document grounded generation tasks: Wikipedia Update Generation task and Dialogue response generation. Our work introduces two novel adaptations of large scale pre-trained encoder-decoder models focusing on building context driven representation of the document and enabling specific attention to the information in the document. Additionally, we provide a stronger BART baseline for these tasks. Our proposed techniques outperform existing methods on both automated (at least 48% increase in BLEU-4 points) and human evaluation for closeness to reference and relevance to the document. Furthermore, we perform comprehensive manual inspection of the generated output and categorize errors to provide insights into future directions in modeling these tasks.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
