Explainable Graph-Based Machine Learning

About

Neural models for relational data provide informative representations for knowledge graphs (KGs), modeling missing links, node classes, and errors in the graph. To aid user trust and provide insights into these models, there has been growing interest in investigating the interpretability and robustness of existing knowledge graph representation models. Previous studies provide explanations for graph models' behavior resulting in designing more accurate and robust models. The objectives of XGML workshop are to bring together researchers interested in (a) explaining graph models, (c) adversarial attacks and defenses on graphs, and (b) improving completion and construction of knowledge bases utilizing the insights from explanations, and in general, to share state-of-the-art approaches, best practices, and future directions.

Speakers


Carolin Lawrence, NEC Lab Europe Title: Explainable Knowledge Graph Prediction for Human-AI Collaboration	Pascal Hitzler, Kansas State University Title: Knowledge Graphs in Neuro-Symbolic AI	Marinka Zitnik, Harvard University Title: Explainable Graph AI for Drug Discovery	Freddy Lecue, CortAIx Title: Explaining Deep Neural Networks: The Good, the Bad and the Ugly

Schedule

XGML workshop will be held virtually on October 8th from 8:25AM - 12:45PM in Pacific Time (UTC-7).

8:25 AM - 8:30 AM : Opening Remarks

8:30 AM - 9:15 AM : Invited talk: Pascal Hitzler, Knowledge Graphs in Neuro-Symbolic AI

9:15 AM - 10:00 AM : Invited talk: Carolin Lawrence, Explainable Knowledge Graph Prediction for Human-AI Collaboration

10:00 AM - 10:15 AM : Contributed talk: Zhepei Wei, CogKG: Unifying Symbolic and Continuous Knowledge Representations for Machine Reasoning

10:15 AM - 10:30 AM : Contributed talk: Xiangyu Peng, Explainable Reinforcement Learning Agent With Stacked Hierarchical Attention

10:30 AM - 10:45 AM : Break

10:45 AM - 11:30 AM : Invited talk: Marinka Zitnik, Explainable Graph AI for Drug Discovery

11:30 AM - 12:15 PM : Invited talk: Freddy Lecue, Explaining Deep Neural Networks: The Good, the Bad and the Ugly

12:15 PM - 12:30 PM : Contributed talk: Joerg Schad, Some Explainable Graph ML Models with ArangoDB

12:30 PM - 12:45 PM : Closing Remarks

CALL FOR PAPERS

Scope

The workshop on Explainable Graph-Based Machine Learning (XGML) will consist of contributed posters, and invited talks on a wide variety of methods and problems in this area. We invite extended abstract submissions in the following categories to present at the workshop:

Providing explanations for different task based on graph models

Reasoning and rule extraction for relational data

Adversarial attacks and defenses

Explainable AI with knowledge graphs

Human in the loop

Optimization challenges in providing explanations

Benchmark datasets and evaluation methods

Call for extended abstract

We invite submission of extended abstracts related to Explainable Graph-Based Machine Learning (XGML). Since the workshop is not intended to have a proceeding comprising full versions of the papers, concurrent submissions to other venues, as well accepted work, are allowed provided that concurrent submissions or intention to submit to other venues is declared to all venues including XGML. Accepted work will be presented as oral during the workshop and listed on this website.

Reviewing Policy:

Submissions shall be refereed on the basis of technical quality, potential impact, and clarity. Atleast one of the authors of each accepted submission will be required to present the work virtually.

Submission instructions

1). Prepare 1-page abstract.
2). Please upload your submission in the following Google form (only PDF accepted):
submission website.
3). In case of any queries, please drop an email to pezeshkp@uci.edu

Important dates

Abstract submission: Aug 20th, 11:59pm PST.
Acceptance/rejection notification: Sep 10th.
Workshop: Oct 8th, 2021.