Research

This page highlights publications from my Google Scholar profile.

Tags: identifiability, symbols, concepts, representations, linear properties, disentanglement, uncertainty, continual

Identifiability and Representation Learning

• Logit distance bounds representational similarity — B. Nielsen, E. Marconato, L. Gresele, A. Dittadi, S. Buchholz. To appear at ICML, 2026.
  representations identifiability linear properties

• When does closeness in distribution imply representational similarity? An identifiability perspective — B. Nielsen, E. Marconato, A. Dittadi, L. Gresele. Advances in Neural Information Processing Systems 38, 102156–102209, 2026.
  representations identifiability

• All or none: Identifiable linear properties of next-token predictors in language modeling — E. Marconato, S. Lachapelle, S. Weichwald, L. Gresele. AISTATS 2025, 2024.
  representations identifiability linear properties

Neuro-Symbolic AI

• Shortcuts and Identifiability in Concept-based Models from a Neuro-Symbolic Lens — S. Bortolotti, E. Marconato, P. Morettin, A. Passerini, S. Teso. Advances in Neural Information Processing Systems 38, 173204–173243, 2026.
  symbols identifiability concepts

• Benchmarking in Neuro-Symbolic AI — A. Bizzarri, A. Borghesi, S. Bortolotti, L. De Raedt, D. Dhami, M. Diligenti, … Learning and Reasoning: 4th International Joint Conference on Learning and …, 2025.
  symbols

• A neuro-symbolic benchmark suite for concept quality and reasoning shortcuts — S. Bortolotti, E. Marconato, T. Carraro, P. Morettin, E. Van Krieken, A. Vergari, … Advances in Neural Information Processing Systems 37, 115861–115905, 2024.
  symbols concepts

• BEARS Make Neuro-Symbolic Models Aware of their Reasoning Shortcuts — E. Marconato, S. Bortolotti, E. van Krieken, A. Vergari, A. Passerini, S. Teso. 40th Conference on Uncertainty in Artificial Intelligence (UAI 2024), 2024.
  symbols uncertainty

• Not all neuro-symbolic concepts are created equal: Analysis and mitigation of reasoning shortcuts — E. Marconato, S. Teso, A. Vergari, A. Passerini. Advances in Neural Information Processing Systems 36, 72507–72539, 2023.
  symbols identifiability

• Neuro symbolic continual learning: Knowledge, reasoning shortcuts and concept rehearsal — E. Marconato, G. Bontempo, E. Ficarra, S. Calderara, A. Passerini, S. Teso. International Conference on Machine Learning (ICML 2023), 2023.
  symbols continual

Concept-based models

• If Concept Bottlenecks are the Question, are Foundation Models the Answer? — N. Debole, P. Barbiero, F. Giannini, A. Passerini, S. Teso, E. Marconato. Machine Learning 115 (5), 97, 2026.
  concepts

• Interpretability is in the mind of the beholder: A causal framework for human-interpretable representation learning — E. Marconato, A. Passerini, S. Teso. Entropy 25 (12), 1574, 2023.
  concepts representations disentanglement

• Glancenets: Interpretable, leak-proof concept-based models — E. Marconato, A. Passerini, S. Teso. Advances in Neural Information Processing Systems 35, 21212–21227, 2022.
  concepts representations disentanglement

• Catastrophic forgetting in continual concept bottleneck models — E. Marconato, G. Bontempo, S. Teso, E. Ficarra, S. Calderara, A. Passerini. International Conference on Image Analysis and Processing, 539–547, 2022.
  concepts continual

More details

For the full list of my publications, see my Google Scholar profile.

—