Call for Papers

Foundation Models have been used successfully in vision and language tasks, yet a large number of real-world applications in finance, healthcare, or industry instead rely on large scale tabular data, which models have largely remained contained to classical machine learning. The emergence of in-context learners applied to tabular data, more precisely Tabular Foundation Models (TFMs), offers a new paradigm that promises to drastically expand the field of applications of tabular data. However, such a paradigm comes with traditional issues of foundation models, such as robustness, privacy, explainability, or fairness. Given the growing interest in TFMs and the potential applications in sensitive domains, we as a community must address the particular issues that come with TFMs.

TrustTFM aims to bring together both the TFMs and structured-data research communities, industrial experts applying TFMs, and the security ML research community to discuss, plan and address the issues in current and future TFMs and in their application. TrustTFM promotes cross-disciplinary discussions and intends to foster and structure a new community along the lines of security related topics in TFMs and their applications.

Submission categories

TrustTFM welcomes submissions from both academia and industry. TrustTFM is a non-proceeding workshop: accepted papers will not be published. The authors retain the rights to submit their work or an extension of their work to any future conference/journal. Only an accepted pre-print version of the paper will be added on the workshop website (with the authors' permission) to foster open research. At least one author from an accepted submission is required to attend the workshop to present the paper.

• Technical paper (up to 6 pages): TrustTFM accepts a large variety of submissions: applied or theoretical methods, empirical studies, industrial case studies, benchmarks/datasets, tools, etc. Papers should detail the trustworthiness problem tackled and the method to address it. Experimental results are not mandatory, but preliminary results are encouraged.

An additional 1-2 pages of references for each submission is allowed. Unlimited appendix if needed. The page limit is strict.

Submission guidelines

Submitted papers must not substantially overlap with papers that have been published or accepted for publication, or that are simultaneously in submission to a journal, conference, or workshop with published proceedings. However, authors may choose to give talks about their work, post a preprint of the paper online, and disclose security vulnerabilities to vendors. Double submission to the workshop and the main conference is not permitted.

Format: Submissions must be a PDF file in two-column IEEE proceeding style. Authors must use

\documentclass[conference]{IEEEtran}

when preparing their paper. Artifacts (e.g. code, demo, etc.) are encouraged but not mandatory.

Review process: Papers will be reviewed by a program committee via double-blind review. As such, papers should be properly anonymized (e.g. remove references to authors' names, institutions, works). Breaking format or anonymity is ground for rejection. There will be no author discussion phase: papers will be accepted or rejected based on the reviews after discussion between the reviewers and, if needed, the PC chairs. Papers will be judged on the novelty of their proposed approach/results, technical soundness and importance of contribution to the workshop topic.