wolfstam / awesome-explanatory-supervision

List of relevant resources for machine learning from explanatory supervision

Awesome Explanatory Supervision

Overview of literature on learning from supervision on the model's explanations.

Warning: early draft, permanent WIP.

Did we miss a relevant paper? Please submit a new entry in the following format:

- **An Artificially-intelligent Means to Escape Discreetly from the Departmental Holiday Party; guide for the socially awkward**
  Eve Armstrong; arXiv 2020 [paper](https://arxiv.org/abs/2003.14169)
  Notes: it is a joke;  a pretty good joke actually.

Table of Contents

• Passive Learning
• Interactive Learning
• Reinforcement Learning
• Distillation
• Regularization without Supervision
• Related Works
• Resources

---

Passive Learning

Approaches that supervise the model's explanations.

• Tangent Prop - A formalism for specifying selected invariances in an adaptive network Patrice Simard, Bernard Victorri, Yann Le Cun, John Denker; NeurIPS 1992 paper Notes: injects invariances into a neural net by regularizing its gradient; precursor to learning from gradient-based explanations.

• Rationalizing Neural Predictions Tao Lei, Regina Barzilay, Tommi Jaakkola; EMNLP 2016 paper Note: they learn an `explanation module' for text classificaiton from explanatory supervision, namely rationales.

• Right for the right reasons: training differentiable models by constraining their explanations Andrew Slavin Ross, Michael C. Hughes, and Finale Doshi-Velez; IJCAI 2017 paper

• Interpretable Machine Teaching via Feature Feedback Shihan Su, Yuxin Chen, Oisin Mac Aodha, Pietro Perona, Yisong Yue; Workshop on Teaching Machines, Robots, and Humans 2017 paper

• Teaching meaningful explanations Noel Codella, Michael Hind, Karthikeyan Ramamurthy, Murray Campbell, Amit Dhurandhar, Kush Varshney, Dennis Wei, Aleksandra Mojsilovic; arXiv 2018 paper

• TED: Teaching AI to explain its decisions Michael Hind, Dennis Wei, Murray Campbell, Noel Codella, Amit Dhurandhar, Aleksandra Mojsilović, Karthikeyan Ramamurthy, Kush Varshney; AIES 2019 paper

• Deriving Machine Attention from Human Rationales Yujia Bao, Shiyu Chang, Mo Yu, and Regina Barzilay; ACL 2019 paper

• Do Human Rationales Improve Machine Explanations? Strout, Julia, Ye Zhang, Raymond Mooney; ACL Workshop BlackboxNLP 2019 paper

• Concept bottleneck models Pang Wei Koh, Thao Nguyen, Yew Siang Tang, Stephen Mussmann, Emma Pierson, Been Kim, and Percy Liang; ICML 2020 paper

• Debiasing Concept Bottleneck Models with Instrumental Variables Mohammad Taha Bahadori, and David E. Heckerman; arXiv 2020 paper

• Learning Global Transparent Models Consistent with Local Contrastive Explanations Tejaswini Pedapati, Avinash Balakrishnan, Karthikeyan Shanmugam, Amit Dhurandhar; NeurIPS 2020 paper

• Reflective-Net: Learning from Explanations Johannes Schneider, Michalis Vlachos; arXiv 2020 paper

• Teaching with Commentaries Aniruddh Raghu, Maithra Raghu, Simon Kornblith, David Duvenaud, and Geoffrey Hinton; arXiv 2020 paper

• Improving performance of deep learning models with axiomatic attribution priors and expected gradients Gabriel Erion, Joseph D. Janizek, Pascal Sturmfels, Scott Lundberg, Su-In Lee; arXiv 2020 paper

• Interpretations are useful: penalizing explanations to align neural networks with prior knowledge Laura Rieger, Chandan Singh, William Murdoch, Bin Yu; ICML 2020 paper

• When Can Models Learn From Explanations? A Formal Framework for Understanding the Roles of Explanation Data Peter Hase, Mohit Bansal; arXiv 2020 paper

• Learning to Faithfully Rationalize by Construction Sarthak Jain, Sarah Wiegreffe, Yuval Pinter, Byron Wallace. ACL 2020 paper code

• Explain and Predict, and then Predict Again Zijian Zhang, Koustav Rudra, Avishek Anand; arXiv 2021 paper

---

Interactive Learning

Approaches that combine supervision on the explanations with interactive machine learning:

• Principles of Explanatory Debugging to Personalize Interactive Machine Learning Todd Kulesza, Margaret Burnett, Weng-Keen Wong, Simone Stumpf; IUI 2015 paper

• Explanatory Interactive Machine Learning Stefano Teso, Kristian Kersting; AIES 2019 paper Note: introduces explanatory interactive learning, focuses on active learning setup.

• Taking a hint: Leveraging explanations to make vision and language models more grounded Ramprasaath R. Selvaraju, Stefan Lee, Yilin Shen, Hongxia Jin, Shalini Ghosh, Larry Heck, Dhruv Batra, and Devi Parikh; ICCV 2019 pdf

• Toward Faithful Explanatory Active Learning with Self-explainable Neural Nets Stefano Teso; IAL Workshop 2019. paper Note: explanatory active learning with self-explainable neural networks.

• Making deep neural networks right for the right scientific reasons by interacting with their explanations Patrick Schramowski, Wolfgang Stammer, Stefano Teso, Anna Brugger, Franziska Herbert, Xiaoting Shao, Hans-Georg Luigs, Anne-Katrin Mahlein, Kristian Kersting; Nature Machine Intelligence 2020 paper Note: introduces end-to-end explanatory interactive learning, fixes clever Hans deep neural nets.

• One explanation does not fit all Kacper Sokol, Peter Flach; 2020 Künstliche Intelligenz paper

• FIND: Human-in-the-loop Debugging Deep Text Classifiers Piyawat Lertvittayakumjorn, Lucia Specia, Francesca Toni; EMNLP 2020 paper

• Human-driven FOL explanations of deep learning Gabriele Ciravegna, Francesco Giannini, Marco Gori, Marco Maggini, Stefano Melacci; IJCAI 2020 paper Notes: first-order logic.

• Machine Guides, Human Supervises: Interactive Learning with Global Explanations Teodora Popordanoska, Mohit Kumar, Stefano Teso; arXiv 2020 paper Note: introduces narrative bias and explanatory guided learning, focuses on human-initiated interaction and global explanations.

• Right for the Right Concept: Revising Neuro-Symbolic Concepts by Interacting with their Explanations Wolfgang Stammer, Patrick Schramowski, and Kristian Kersting; arXiv 2020 paper Notes: first-order logic, attention.

• Right for Better Reasons: Training Differentiable Models by Constraining their Influence Function Xiaoting Shao, Arseny Skryagin, Patrick Schramowski, Wolfgang Stammer, Kristian Kersting; AAAI 2021 preliminary paper

• Bandits for Learning to Explain from Explanations Freya Behrens, Stefano Teso, Davide Mottin; XAI Workshop 2021 paper Notes: preliminary.

• Cost-effective Interactive Attention Learning with Neural Attention Process Jay Heo, Junhyeon Park, Hyewon Jeong, Kwang joon Kim, Juho Lee, Eunho Yang, Sung Ju Hwang; ICML 2020 paper Notes: attention, interaction

---

Reinforcement Learning

• Explanation Augmented Feedback in Human-in-the-Loop Reinforcement Learning Lin Guan, Mudit Verma, Sihang Guo, Ruohan Zhang, Subbarao Kambhampati; arXiv 2020 paper

---

Distillation

• Model reconstruction from model explanations Smitha Milli, Ludwig Schmidt, Anca D. Dragan, Moritz Hardt; FAcct 2019 paper

• Evaluating Explanations: How much do explanations from the teacher aid students? Danish Pruthi, Bhuwan Dhingra, Livio Baldini Soares, Michael Collins, Zachary C. Lipton, Graham Neubig, and William W. Cohen; arXiv 2020 paper Notes: defines importance of different kinds of explanations by measuring their impact when used as supervision.

---

Regularization without Supervision

Approaches that regularize the model's explanations in an unsupervised manner, often for improved interpretability.

• Towards robust interpretability with self-explaining neural networks David Alvarez-Melis, Tommi Jaakkola; NeurIPS 2018 paper

• Beyond sparsity: Tree regularization of deep models for interpretability Mike Wu, Michael Hughes, Sonali Parbhoo, Maurizio Zazzi, Volker Roth, Finale Doshi-Velez; AAAI 2018 paper

• Regional tree regularization for interpretability in deep neural networks Mike Wu, Sonali Parbhoo, Michael Hughes, Ryan Kindle, Leo Celi, Maurizio Zazzi, Volker Roth, Finale Doshi-Velez; AAAI 2020 paper

• Regularizing black-box models for improved interpretability Gregory Plumb, Maruan Al-Shedivat, Ángel Alexander Cabrera, Adam Perer, Eric Xing, Ameet Talwalkar; NeurIPS 2020 paper

• Explainable Models with Consistent Interpretations Vipin Pillai, Hamed Pirsiavash; AAAI 2021 paper code

---

Related Works

Explanation-based learning, focuses on logic-based formalisms and learning strategies:

• Explanation-based generalization: A unifying view Tom Mitchell, Richard Keller, Smadar Kedar-Cabelli; MLJ 1986 paper

• Explanation-based learning: An alternative view Gerald DeJong, Raymond Mooney; MLJ 1986 paper

• Explanation-based learning: A survey of programs and perspectives Thomas Ellman; ACM Computing Surveys 1989 paper

• Probabilistic explanation based learning Angelika Kimmig, Luc De Raedt, Hannu Toivonen; ECML 2007 paper

Injecting invariances / feature constraints into models:

• Training invariant support vector machines Dennis DeCoste, Bernhard Schölkopf; MLJ 2002 paper

• The constrained weight space svm: learning with ranked features Kevin Small, Byron Wallace, Carla Brodley, Thomas Trikalinos; ICML 2011 paper

Dual label-feature feedback:

• Active learning with feedback on features and instances Hema Raghavan, Omid Madani, Rosie Jones; JMLR 2006 paper

• An interactive algorithm for asking and incorporating feature feedback into support vector machines Hema Raghavan, James Allan; ACM SIGIR 2007 paper

• Learning from labeled features using generalized expectation criteria Gregory Druck, Gideon Mann, Andrew McCallum; ACM SIGIR 2008 paper

• Active learning by labeling features Gregory Druck, Burr Settles, Andrew McCallum; EMNLP 2009 paper

• A unified approach to active dual supervision for labeling features and examples Josh Attenberg, Prem Melville, Foster Provost; ECML-PKDD 2010 paper

• Closing the loop: Fast, interactive semi-supervised annotation with queries on features and instances Burr Settles; EMNLP 2011 paper

Learning from rationales:

• Using “annotator rationales” to improve machine learning for text categorization Omar Zaidan, Jason Eisner, Christine Piatko; NAACL 2007 paper

• Modeling annotators: A generative approach to learning from annotator rationales Omar Zaidan, Jason Eisner; EMNLP 2008 paper

• Active learning with rationales for text classification Manali Sharma, Di Zhuang, Mustafa Bilgic; NAACL 2015 paper

Critiquing in recommenders:

• Critiquing-based recommenders: survey and emerging trends Li Chen, Pearl Pu; User Modeling and User-Adapted Interaction 2012 paper

• Coactive critiquing: Elicitation of preferences and features Stefano Teso, Paolo Dragone, Andrea Passerini; AAAI 2017 paper

---

Resources

A selection of general resources on Explainable AI focusing on overviews, surveys, societal implications, and critiques:

• Survey and critique of techniques for extracting rules from trained artificial neural networks Robert Andrews, Joachim Diederich, Alan B. Tickle; Knowledge-based systems 1995 page

• The Mythos of Model Interpretability Zachary Lipton; CACM 2016 paper

• A survey of methods for explaining black box models Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri, Franco Turini, Fosca Giannotti, and Dino Pedreschi; ACM Computing Surveys 2018 paper

• Explanation in Artificial Intelligence: Insights from the Social Sciences Tim Miller; Artificial Intelligence, 2019 paper

• Unmasking clever hans predictors and assessing what machines really learn Sebastian Lapuschkin, Stephan Wäldchen, Alexander Binder, Grégoire Montavon, Wojciech Samek, Klaus-Robert Müller; Nature Communications 2019 paper and

• Interpretation of neural networks is fragile Amirata Ghorbani, Abubakar Abid, James Zou; AAAI 2019 paper

• Is Attention Interpretable? Sofia Serrano, Noah A. Smith; ACL 2019 paper

• Attention is not Explanation Sarthak Jain, and Byron C. Wallace; ACL 2019 paper

• Attention is not not Explanation Sarah Wiegreffe, and Yuval Pinter; EMNLP-IJCNLP 2019 paper

• The (un)reliability of saliency methods Pieter-Jan Kindermans, Sara Hooker, Julius Adebayo, Maximilian Alber, Kristof T. Schütt, Sven Dähne, Dumitru Erhan, and Been Kim; Explainable AI: Interpreting, Explaining and Visualizing Deep Learning 2019 paper

• Explanations can be manipulated and geometry is to blame Ann-Kathrin Dombrowski, Maximillian Alber, Christopher Anders, Marcel Ackermann, Klaus-Robert Müller, and Pan Kessel; NeurIPS 2019 paper

• Fooling Neural Network Interpretations via Adversarial Model Manipulation Juyeon Heo, Sunghwan Joo, and Taesup Moon; NeurIPS 2019 paper

• Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead Cynthia Rudin; Nature Machine Intelligence 2019 page

• Shortcut learning in deep neural networks. Robert Geirhos, Jorn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, Felix Wichmann; Nature Machine Intelligence 2020 page

---

Related Lists

• Awesome explainable AI

• Awesome machine learning interpretability

---

Not Yet Sorted

• e-SNLI: natural language inference with natural language explanations Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom; NeurIPS 2018 paper

• Multimodal explanations: Justifying decisions and pointing to the evidence Dong Huk Park, Lisa Anne Hendricks, Zeynep Akata, Anna Rohrbach, Bernt Schiele, Trevor Darrell, Marcus Rohrbach; CVPR 2018 paper

• Learning Deep Attribution Priors Based On Prior Knowledge Ethan Weinberger, Joseph Janizek, Su-In Lee; NeurIPS 2020 paper

---

TODO

• Make sure that all papers are categorized correctly ;-)

• Add link to code wherever available.

• Crawl & reference work on NLP.

Comments

This list is directly inspired by all the awesome awesome lists out there!