اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدAttention is Not Only a Weight: Analyzing Transformers with Vector Norms
148
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Attention is a key component of Transformers, which have recently achieved considerable success in natural language processing. Hence, attention is being extensively studied to investigate various linguistic capabilities of Transformers, focusing on analyzing the parallels between attention weights and specific linguistic phenomena. This paper shows that attention weights alone are only one of the two factors that determine the output of attention and proposes a norm-based analysis that incorporates the second factor, the norm of the transformed input vectors. The findings of our norm-based analyses of BERT and a Transformer-based neural machine translation system include the following: (i) contrary to previous studies, BERT pays poor attention to special tokens, and (ii) reasonable word alignment can be extracted from attention mechanisms of Transformer. These findings provide insights into the inner workings of Transformers.
قيم البحث
اقرأ أيضاً
Attention mechanisms play a central role in NLP systems, especially within recurrent neural network (RNN) models. Recently, there has been increasing interest in whether or not the intermediate representations offered by these modules may be used to
explain the reasoning for a models prediction, and consequently reach insights regarding the models decision-making process. A recent paper claims that `Attention is not Explanation (Jain and Wallace, 2019). We challenge many of the assumptions underlying this work, arguing that such a claim depends on ones definition of explanation, and that testing it needs to take into account all elements of the model, using a rigorous experimental design. We propose four alternative tests to determine when/whether attention can be used as explanation: a simple uniform-weights baseline; a variance calibration based on multiple random seed runs; a diagnostic framework using frozen weights from pretrained models; and an end-to-end adversarial attention training protocol. Each allows for meaningful interpretation of attention mechanisms in RNN models. We show that even when reliable adversarial distributions can be found, they dont perform well on the simple diagnostic, indicating that prior work does not disprove the usefulness of attention mechanisms for explainability.
We study the power of cross-attention in the Transformer architecture within the context of transfer learning for machine translation, and extend the findings of studies into cross-attention when training from scratch. We conduct a series of experime
nts through fine-tuning a translation model on data where either the source or target language has changed. These experiments reveal that fine-tuning only the cross-attention parameters is nearly as effective as fine-tuning all parameters (i.e., the entire translation model). We provide insights into why this is the case and observe that limiting fine-tuning in this manner yields cross-lingually aligned embeddings. The implications of this finding for researchers and practitioners include a mitigation of catastrophic forgetting, the potential for zero-shot translation, and the ability to extend machine translation models to several new language pairs with reduced parameter storage overhead.
With the increasing empirical success of distributional models of compositional semantics, it is timely to consider the types of textual logic that such models are capable of capturing. In this paper, we address shortcomings in the ability of current
models to capture logical operations such as negation. As a solution we propose a tripartite formulation for a continuous vector space representation of semantics and subsequently use this representation to develop a formal compositional notion of negation within such models.
Numerous papers ask how difficult it is to cluster data. We suggest that the more relevant and interesting question is how difficult it is to cluster data sets {em that can be clustered well}. More generally, despite the ubiquity and the great import
ance of clustering, we still do not have a satisfactory mathematical theory of clustering. In order to properly understand clustering, it is clearly necessary to develop a solid theoretical basis for the area. For example, from the perspective of computational complexity theory the clustering problem seems very hard. Numerous papers introduce various criteria and numerical measures to quantify the quality of a given clustering. The resulting conclusions are pessimistic, since it is computationally difficult to find an optimal clustering of a given data set, if we go by any of these popular criteria. In contrast, the practitioners perspective is much more optimistic. Our explanation for this disparity of opinions is that complexity theory concentrates on the worst case, whereas in reality we only care for data sets that can be clustered well. We introduce a theoretical framework of clustering in metric spaces that revolves around a notion of good clustering. We show that if a good clustering exists, then in many cases it can be efficiently found. Our conclusion is that contrary to popular belief, clustering should not be considered a hard task.
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.
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
