Hosting offer for MSCA-PF-2026 candidates

We are looking for enthusiastic, independent researchers to apply for a Marie Skłodowska-Curie Postdoctoral Fellowship (MSCA-PF-2026) hosted in our group. The fellowship is co-written by the researcher and the host, and we provide hands-on support throughout. We welcome candidates in one of two areas:

1. Responsible AI: We want to make LLMS, VLMS and beyond — trustworthy, transparent, and accountable. Our priority directions are:

• Interpretability of multimodal AI : understanding how multimodal models represent and combine information across modalities; mechanistic interpretability, cross-modal attribution, and faithful explanations of model behaviour.

• Fairness : auditing and mitigating bias in multimodal models, including how bias arises and compounds across image, text, and other modalities.

• Machine unlearning : efficiently removing the influence of specific data or knowledge from trained multimodal models to support privacy, the right to be forgotten, and regulatory compliance; exact vs. approximate unlearning, and verifying what has actually been forgotten.

• Protocol learning : decentralized, communication- and energy-efficient collaborative training coordinated by a protocol rather than a central server, with contribution-based incentives and governance built in.

Work aligning these directions with emerging standards such as the EU AI Act is especially welcome.

2. Quantum Machine Learning. Variational quantum algorithms and quantum neural networks, quantum kernels and data-encoding strategies, trainability (e.g. barren plateaus), hybrid quantum–classical models, quantum generative models, and error mitigation for NISQ-era hardware — as theory/algorithms or application-driven work.

What we offer. Hands-on support in developing a competitive proposal — shaping the research idea, co-writing the scientific sections, and preparing for the three MSCA evaluation criteria (Excellence, Impact, Implementation) — plus access to our group's expertise, infrastructure, and collaboration network.