An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction

Paranjape, Bhargavi; Joshi, Mandar; Thickstun, John; Hajishirzi, Hannaneh; Zettlemoyer, Luke

doi:10.48550/arxiv.2005.00652

Cited by 2 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Incorporating the information bottleneck theory [37,138]. The reduction of the original input text inevitably results in performance degradation because the rationale complement usually contains non-essential but useful information.…”

Section: Representative Workmentioning

confidence: 99%

See 1 more Smart Citation

Interpreting Deep Learning Models in Natural Language Processing: A Review

Sun¹,

Yang²,

Li³

et al. 2021

Preprint

View full text Add to dashboard Cite

Neural network models have achieved state-of-the-art performances in a wide range of natural language processing (NLP) tasks. However, a long-standing criticism against neural network models is the lack of interpretability, which not only reduces the reliability of neural NLP systems but also limits the scope of their applications in areas where interpretability is essential (e.g., health care applications). In response, the increasing interest in interpreting neural NLP models has spurred a diverse array of interpretation methods over recent years. In this survey, we provide a comprehensive review of various interpretation methods for neural models in NLP. We first stretch out a high-level taxonomy for interpretation methods in NLP, i.e., training-based approaches, test-based approaches and hybrid approaches. Next, we describe sub-categories in each category in detail, e.g., influence-function based methods, KNN-based methods, attention-based models, saliency-based methods, perturbation-based methods, etc. We point out deficiencies of current methods and suggest some avenues for future research.

show abstract

Section: Representative Workmentioning

confidence: 99%

“…Figure 16: Diagrams of the information bottleneck enhanced model (left, the figure is brought from [138]) and the distribution matching model (right, the figure is brought from [79]). Left: An explainer first extracts rationales from the input and then a classifier predicts the output based only on the rationale.…”

Section: Representative Workmentioning

confidence: 99%

Interpreting Deep Learning Models in Natural Language Processing: A Review

Sun¹,

Yang²,

Li³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Yu et al (2019) employed a similar idea and additionally employed constraints on the complement of rationales. Other approaches were based on the information bottleneck (Paranjape et al 2020), latent variable models (Bastings, Aziz, and Titov 2019), and learning environment-invariant representations (Chang et al 2020). However, none of these previous methods consider the relationship between rationales and the full input in terms of feature distribution.…”

Section: Related Work Interpretabilitymentioning

confidence: 99%

Distribution Matching for Rationalization

Huang

Chen²,

Du³

et al. 2021

AAAI

View full text Add to dashboard Cite

The task of rationalization aims to extract pieces of input text as rationales to justify neural network predictions on text classification tasks. By definition, rationales represent key text pieces used for prediction and thus should have similar classification feature distribution compared to the original input text. However, previous methods mainly focused on maximizing the mutual information between rationales and labels while neglecting the relationship between rationales and input text. To address this issue, we propose a novel rationalization method that matches the distributions of rationales and input text in both the feature space and output space. Empirically, the proposed distribution matching approach consistently outperforms previous methods by a large margin. Our data and code are available.

show abstract

An Information Bottleneck Approach for Controlling Conciseness in Rationale Extraction

Cited by 2 publications

References 0 publications

Interpreting Deep Learning Models in Natural Language Processing: A Review

Interpreting Deep Learning Models in Natural Language Processing: A Review

Distribution Matching for Rationalization

Contact Info

Product

Resources

About