Proceedings Article10.48550/arXiv.2205.03286
GlobEnc: Quantifying Global Token Attribution by Incorporating the Whole Encoder Layer in Transformers
A. Modarressi,Mohsen Fayyaz,Yadollah Yaghoobzadeh,Mohammad Taher Pilehvar +3 more
- 06 May 2022
Vol. abs/2205.03286
22
TL;DR: A novel token attribution analysis method that incorporates all the components in the encoder block and aggregates this throughout layers and significantly outperforms previous methods on various tasks regarding correlation with gradient-based saliency scores.
read more
Abstract: There has been a growing interest in interpreting the underlying dynamics of Transformers. While self-attention patterns were initially deemed as the primary option, recent studies have shown that integrating other components can yield more accurate explanations. This paper introduces a novel token attribution analysis method that incorporates all the components in the encoder block and aggregates this throughout layers. Through extensive quantitative and qualitative experiments, we demonstrate that our method can produce faithful and meaningful global token attributions. Our experiments reveal that incorporating almost every encoder component results in increasingly more accurate analysis in both local (single layer) and global (the whole model) settings. Our global attribution analysis significantly outperforms previous methods on various tasks regarding correlation with gradient-based saliency scores. Our code is freely available at https://github.com/mohsenfayyaz/GlobEnc.
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Citations
From Understanding to Utilization: A Survey on Explainability for Large Language Models
TL;DR: This survey underscores the imperative for increased explainability in LLMs, delving into both the research on explainability and the various methodologies and tasks that utilize an understanding of these models.
16
Quantifying Context Mixing in Transformers
Hosein Mohebbi,Willem Zuidema,Grzegorz Chrupała,Afra Alishahi +3 more
- 30 Jan 2023
TL;DR: The authors propose Value Zeroing, a context mixing score customized for Transformers that provides a deeper understanding of how information is mixed at each encoder layer, and demonstrate the superiority of their context mixing scores over other analysis methods through a series of complementary evaluations with different viewpoints based on linguistically informed rationales, probing, and faithfulness analysis.
14
Feed-Forward Blocks Control Contextualization in Masked Language Models
01 Feb 2023
TL;DR: This article analyzed the inner contextualization of Transformer-based models by considering all the components, including the feed-forward block and its surrounding residual and normalization layers, as well as the attention.
DecompX: Explaining Transformers Decisions by Propagating Token Decomposition
A. Modarressi,Mohsen Fayyaz,Ehsan Aghazadeh,Yadollah Yaghoobzadeh,Mohammad Taher Pilehvar +4 more
- 05 Jun 2023
TL;DR: DecompX as mentioned in this paper is based on the construction of decomposed token representations and their successive propagation throughout the model without mixing them in between layers, which provides multiple advantages over existing solutions for its inclusion of all encoder components and the classification head.
Establishing An Optimal Online Phishing Detection Method: Evaluating Topological NLP Transformers on Text Message Data
Helen Milner
- 12 Jul 2023
TL;DR: In this article , an optimal classification model for online SMS spam detection by utilizing topological sentence transformer methodologies was established. Butler et al. presented a viable lightweight integration of pre-trained NLP repository models with sklearn functionality, which achieved an optimal F1-score of 0.938.
5
References
•Proceedings Article
Attention is All you Need
Ashish Vaswani,Noam Shazeer,Niki Parmar,Jakob Uszkoreit,Llion Jones,Aidan N. Gomez,Lukasz Kaiser,Illia Polosukhin +7 more
- 12 Jun 2017
TL;DR: This paper proposed a simple network architecture based solely on an attention mechanism, dispensing with recurrence and convolutions entirely and achieved state-of-the-art performance on English-to-French translation.
•Posted Content
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
TL;DR: A new language representation model, BERT, designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers, which can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks.
81.7K
•Proceedings Article
Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank
Richard Socher,Alex Perelygin,Jean Y. Wu,Jason Chuang,Christopher D. Manning,Andrew Y. Ng,Christopher Potts +6 more
- 01 Oct 2013
TL;DR: A Sentiment Treebank that includes fine grained sentiment labels for 215,154 phrases in the parse trees of 11,855 sentences and presents new challenges for sentiment compositionality, and introduces the Recursive Neural Tensor Network.
•Proceedings Article
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
Karen Simonyan,Andrea Vedaldi,Andrew Zisserman +2 more
- 23 Dec 2013
TL;DR: In this paper, the gradient of the class score with respect to the input image is computed to compute a class saliency map, which can be used for weakly supervised object segmentation using classification ConvNets.
•Posted Content
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
TL;DR: This work proposes a method to pre-train a smaller general-purpose language representation model, called DistilBERT, which can be fine-tuned with good performances on a wide range of tasks like its larger counterparts, and introduces a triple loss combining language modeling, distillation and cosine-distance losses.
7.3K