================================================================================
CALL FOR PAPERS

MFMB 2026 - First Workshop on Multimodal and Foundation Models in Banking
A workshop at ACM ICAIF '26, Bocconi University, Milan, Italy

Workshop: 14 or 15 November 2026 (exact day to be confirmed by ICAIF)
Website: https://thebal.ai/mfmb2026/
Submit: https://cmt3.research.microsoft.com/MFMB2026

Submission deadline: 2 October 2026, 23:59 AoE
================================================================================


IMPORTANT DATES
---------------
Submission site opens 13 August 2026
Paper submission deadline 2 October 2026 (23:59 AoE)
Author notification 14 October 2026
Camera-ready deadline 30 October 2026
Workshop 14 or 15 November 2026
ICAIF '26 main conference 14-17 November 2026


ABOUT THE WORKSHOP
------------------
The AI landscape has shifted rapidly towards multimodality and foundation
models. Combining tabular, time-series, text and image data in one multimodal
model can predict more accurately than a model using only one of these sources,
solve previously unsolvable problems, and bring competitive advantages. A single
foundation model trained on transactional data can learn general customer
behaviour and then be applied to credit scoring, fraud detection, or pricing.

Deploying multimodal and foundation models in banking is an end-to-end,
multi-stage problem, not only a model-architecture problem. The technical and
methodological questions at each stage are research-worthy and are currently
scattered across disparate venues. MFMB convenes researchers and practitioners
working across these stages around one question: what does rigorous research
methodology look like when multimodal and foundation models meet consequential
banking decisions?

The workshop is organised around six stages:

1. Representation learning over heterogeneous financial data
2. Multimodal architecture design and fusion strategies
3. Evaluation under deployment conditions
4. Counterfactual reasoning under endogenous data
5. Explanation and governance under regulatory obligation
6. Simulation and synthetic data


TOPICS OF INTEREST
------------------
Topics include, but are not limited to:

* Architectures and fusion strategies. Novel neural architectures for
integrating heterogeneous financial data, for example aligning customer
transaction histories with text from application forms, credit-bureau
records, and identity documents.

* Financial foundation models. Pre-training objectives, scaling laws, and
transfer-learning techniques for tabular, transactional, credit, or market
data.

* Trustworthy AI and regulation. Interpretability, algorithmic fairness, and
robust governance in compliance with global financial regulations
(EU AI Act, Basel, IFRS 9).

* Efficient adaptation. Parameter-efficient fine-tuning and distillation of
massive models for low-latency financial environments.

* Privacy and security. Federated learning and differential privacy for
sensitive multi-party financial datasets.

* Applications. Credit scoring, high-frequency fraud detection, automated
auditing, and generative macroeconomic modelling.

* Evaluation under deployment conditions. Out-of-time evaluation protocols,
selection-bias-aware metrics, segment-conditional performance, calibration
under operating thresholds, statistical rigour for rare-label scoring, and
the benchmark-deployment evidence gap.

* Counterfactual reasoning and off-policy evaluation. Reject inference,
policy-change estimation, treatment-effect estimation for interventions
(collection strategies, servicing offers, credit-line changes), adverse
action explanations, and the interaction between foundation-model
representations and causal identification.

* Simulation and synthetic data. Calibrated simulators and synthetic data for
model improvement on under-represented segments and rare events;
simulator-based stress testing under counterfactual regimes; validation
methodology for simulator fidelity.

In scope on the application side: consumer credit underwriting; fraud detection
(transaction, identity, application, account takeover); collections and
recovery; servicing decisions; identity verification and KYC; payment
authorization; BNPL; insurance underwriting where data shapes are similar; and
small-business credit.

On the model side: tabular foundation models, event-sequence transformers,
multimodal fusion architectures, hybrid representations, and other
foundation-model architectures applied to these domains.


SUBMISSION INSTRUCTIONS
-----------------------
Length: Extended abstracts and short papers, maximum 4 pages excluding
references.

Format: ACM sigconf two-column template, consistent with the ICAIF '26 main
track. Use the sigconf class with the anonymous parameter.
LaTeX template: https://portalparts.acm.org/hippo/latex_templates/acmart-primary.zip
Overleaf: https://www.overleaf.com/gallery/tagged/acm-official

Review: Double blind. Every paper is reviewed by two expert reviewers drawn
from the workshop programme committee. Submissions must not reveal
the authors' identities, either directly or through obvious
reference to previous work. Cite your own work in the third person.

Portal: https://cmt3.research.microsoft.com/MFMB2026
MFMB runs its own CMT installation, separate from the ICAIF '26
main track.

NON-ARCHIVAL: MFMB is a non-archival venue. Accepted papers do not appear in
the ACM proceedings, and presenting at the workshop does not preclude
publishing the same work elsewhere, before or after the workshop. Work that has
already appeared as a preprint or in another non-archival venue is welcome.

Accepted papers are posted on the workshop website only with the explicit
authorisation of the authors. Authors are asked for that permission after
notification; consent is opt-in and may be withdrawn at any time. Where it is
not given, only the paper's title and authors are listed.


FORMAT AND PROGRAMME
--------------------
MFMB is a highly interactive half-day workshop combining an invited keynote,
peer-reviewed oral and poster sessions, and a concluding panel discussion.
Accepted papers are presented as a 20-minute contributed talk with 10 minutes
of Q&A, or in the poster session.

Invited keynote:
Tadas Krisciunas, Head of Credit Data Science, Revolut
Anton Repushko, Head of Research, Revolut
A joint keynote from the team implementing tabular foundation models live in
production credit decisioning.

Closing panel:
"Bridging the gap between tabular benchmarks and deployment"


ORGANISING COMMITTEE
--------------------
Maria Oskarsdottir, University of Southampton & Reykjavik University
(primary contact)
Amey Baokar, Revolut
Cristian Bravo, Western University
Stefan Lessmann, Humboldt University of Berlin
Anton Repushko, Revolut
Alexander Statnikov, Affirm


CONTACT
-------
Maria Oskarsdottir, m.oskarsdottir@soton.ac.uk
Workshop website: https://thebal.ai/mfmb2026/