The AAAI-22 Tutorial on

Explanations in Interactive Machine Learning

GOAL

This tutorial is intended for Artificial Intelligence researchers and practitioners, as well as domain experts interested in human-in-the-loop machine learning, including interactive recommendation and active learning.

The participants will gain an understanding of current developments in interactive machine learning from rich human feedback – with an emphasis on white-box interaction and explanation-guided learning – as well as a conceptual map of the variety of methods available and of the relationships between them.

The main goal is to inform the audience about the state-of-the-art in explanations for interactive machine learning, open issues and research directions, and how these developments relate to the broader context of machine learning and AI.

OUTLINE & MATERIAL

The tutorial is split into four main sessions, of roughly 22 minutes each, followed by a Q&A:

Welcome (5min).
Motivation and Challenges (22min), presented by Öznur Alkan.
Interacting via Local Explanations (22min), presented by Stefano Teso.
Interacting via Global Explanations (22min), presented by Elizabeth Daly.
Interaction via Concept-based Explanations (22min), presented by Wolfgang Stammer.
Q&A (12min).

RECORDING & SLIDES

tutorial.mp4

Slides: PDF

PREREQUISITES

Basic knowledge of machine learning at the level of an introductory course is required. An elementary understanding of active learning and recommender systems will be assumed. All other material required will be covered during the tutorial.

Speakers

Öznur Alkan, Optum-United Health Group

Bio: Öznur is a Lead AI Engineer at Optum-United Health Group. [Google Scholar] Öznur’s research focuses on exploring different collaboration techniques between machine learning systems and the user, designing solutions that can facilitate this collaboration in the areas of structured predictive models and recommender systems. She is serving as a PC/SPC at international conferences such as AAAI, IJCAI, CIKM, IUI, UMAP and RecSys.

Stefano Teso, University of Trento

Bio: Assistant Professor at the University of Trento. [Google Scholar] Stefano's work focuses on interactive machine learning, explainable AI, preference elicitation, structured prediction, and con straint learning. Stefano co-organized tutorials on constraint learning at AAAI’18, IJCAI’18, ECML-PKDD’19, and in 2020 he co-edited the special issue on interactive machine learning of the AI journal K ̈unstliche Intelligenz, Springer. He is a regular PC/SPC at AAAI, IJCAI, NeurIPS, ICML, ICLR, ECML-PKDD, UAI and an AC at AISTATS.

Elizabeth Daly, IBM Ireland

Bio: Elizabeth is STSM at IBM Research, Ireland. [Google Scholar] Elizabeth’s work focuses on innovative solutions for interactive AI where systems influence users and users influence systems. She is on the program committee of conferences such as RecSys, IUI, WWW, UMAP and ICWSM. Elizabeth has served as Program Chair, Workshop Chair and Demo Chair for RecSys and organised workshops at IUI, ICWSM and CSCW. She is currently serving on the Royal Irish Academy’s committee on Engineering and Computer Science.

Wolfgang Stammer, Technical University Darmstadt

Bio: Wolfgant is a Ph.D. Student at the Technical University of Darmstadt. [Google Scholar] Wolfgang’s work focuses on explainable and interactive learning as well as neuro-symbolic models and discrete neural representations for improved reasoning and human-machine interactions. He has been a reviewer for international conferences such as AAAI and NeurIPS.

Page updated

Google Sites

Report abuse