What Is Explainability And A Thorough Explanation Of The Current State Of Explainable AI In NLP!
3 main points
✔️ On Explainable AI in Natural Language Processing
✔️ On the current state of explainable natural language processing models
✔️ The challenges of achieving accountability
A Survey of the State of Explainable AI for Natural Language Processing
written by Marina Danilevsky, Kun Qian, Ranit Aharonov, Yannis Katsis, Ban Kawas, Prithviraj Sen
(Submitted on 1 Oct 2020)
Comments: Accepted at AACL-IJCNLP2020
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Introduction
Deep learning models, which have enjoyed great success in recent years, face the problem of difficulty in interpreting the criteria for their models (they are black boxes).
This lack of accountability can also undermine the credibility of the model with users.
For this reason, research on Explainable AI (XAI ), which reveals the reasons/rationale for decisions of deep learning models, has been gaining importance in recent years. Although this research on Explainable AI is a very important subject, it is a new area of research and therefore difficult to gain systematic knowledge.
In light of this lack of information, this article summarizes the current state of Explainable AI in natural language processing. If you are interested in Explainable AI, please take a look at this article.
table of contents
1. Broad Classification of Explainable AI
Local/Global
Self-explanation/post-explanation
2. details of the technology used for explainable AI
2.1. Explainability techniques
Feature importance
Surrogate model
Example-driven
Provenance-based
Declarative induction
2.2. Relevant Technologies of Accountability
First-derivative saliency
Layer-wise relevance propagation
Input perturbations
Attention
LSTM gating signals
Explainability-aware architecture design
2.3. Explainability Visualization Techniques
Saliency
Raw declarative representations
Natural language explanation
3. Assessing Explainability
3.1. Classification of the evaluation
An informal examination of explanations
Comparing to ground truth
Human evaluation
Other
3.2. what is explicable?
4. Summary of actual research cases
To read more,
Please register with AI-SCHOLAR.
OR
Categories related to this article
