Call for Paper: Mini-track on Trustworthy Artificial Intelligence and Machine Learning

as part of Hawaii International Conference on Systems Sciences (HICSS)

Hilton Waikoloa Village on January 7-10, 2025 (Big Island)

This HICSS mini-track is part of the research track: Software Technology

https://hicss.hawaii.edu/tracks-58/software-technology/

With the advancement of AI technology, AI algorithms start to match human performance for certain tasks (e.g. ChatGPT) and discover loopholes in systems that were not previously found.  AI in general and ML methods specifically are increasingly used with scientific data and applied with great promise to solve a broad variety of scientific problems.  With the increased use of AI comes an increase in inherent complexity.  Deep Learning (DL) models with billions of parameters, operating with very large data volumes on heterogeneous architectures, obscure their inner workings to human understanding.  Unlike traditional ML algorithms, such as rule-based decision trees or linear-regression models where the decision boundary is clear, interpreting a learned model is difficult.

The increased need for transparency is compounded by that of avoiding bias in predictions.  Numerous examples of bias have been discovered in image recognition, classification, and text generation. Formal explanations of how models achieve results, explicit representations of data, comprehensiveness and diversity of datasets used for training are crucial to foster trust in AI. Additionally, the accuracy of results obtained with AI is often the product of customization; experiments show that many are not reproducible at scale, even within expected error bounds. While reproducibility may not be needed for some uses of AI (e.g. when AI is used for the purpose of preliminary triage in drug discovery) in other uses, reproducible AI is paramount.    

Researchers need to understand how Artificial Intelligence (AI) and Machine Learning (ML) results are obtained in order to gain new insights and to establish confidence in the validity of these results. The promises of AI/ML will not be realized if scientists cannot trust the results, understand how they were obtained, gain transparency into what datasets, models and model parameters have been used or what features in the data lead to results. Like any good scientific results, AI/ML pipelines should be reproducible to the most possible extent. 

The Trustworthy Artificial Intelligence and Machine Learning mini-track at HICSS 57 will explore a number of themes related to explainable, reproducible, ethical, and trustworthy AI.  Papers will be published with conference proceedings.  Topics of interest include but are not limited to the following:

• Computational and foundational methods for explaining AI models (XAI)

• Computational and foundational methods to ensure reproducibility of machine learning predictions

• Approaches discussing how Findable, Accessible, Interpretable, Re-usable (FAIR) principles can be applied to AI/ML including use of methods

• Mental models for interpreting AI results

• Computational and foundational methods and algorithms for detecting bias in AI/ML models

• Approaches, tools and best practices to perform sanity checks on data transformation pipelines

• Approaches, tools and best practices to keep track of experiments

• Use cases of explainability and reproducibility with AI models (XAI, RAI)

• Case studies of when AI models introduce bias in results

• Definitions and examples of trustworthy AI

The HICSS conferences are system sciences conferences; they are uniquely positioned to propose research directions and bring some answers to the societal questions of trustworthy, explainable artificial intelligence. In its 58th edition, the HICSS conference tackles a breath of topics organized in tracks and mini-tracks, including this one.

We are expecting full papers, 10 pages maximum.  Participants interested  in submitting short papers such as abstracts or position papers should contact the co-chairs.   At HICSS there is room for such papers in our companion tutorial on FAIR and reproducibility.  Practice papers and case studies are welcome.  Selected papers will be published as parts of the conference proceedings and extended versions will be selected for a journal special issue.

Important Dates for Paper Submission  (TBA 2024 - 2023 dates given below as indication)
June 15, 2024 | 11:59 pm HST       Submission Deadline
August 17, 2024 | 11:59 pm HST:        Notification of Acceptance/Rejection
September 22, 2024|11:59 pm HST:   Deadline for Submission of Final Manuscript for Publication
October 1, 2024 | 11:59 pm HST:        Deadline for at least one author to register for HICSS-57

January 7-10, 2025:                           HICSS conference

Submission:  https://hicss-submissions.org/

Author guidelines:  https://hicss.hawaii.edu/authors/

Program Committee: Per HICSS policy, paper reviews in mini-tracks are double-blind.  Names of the program committee members are withdrawn.

Minitrack Co-Chairs:

Line Pouchard, DOE Brookhaven National Laboratory, pouchard@bnl.gov

Main point of contact.

Peter Salhofer

FH JOANNEUM University of Applied Sciences, peter.salhofer@fh-joanneum.at

If you are interested in FAIR guiding principles, see also the 3rd FAIR tutorial at HICSS 57 https://sites.google.com/view/data-stewardship-and-reuse/home