• Training machines to communicate, interact and benefit humanity.

News

The Role of AI in Education: Let Us Help Students Make the Right Tech Decisions

Now that most students are using generative AI in any form to deliver essays, code and more, educators and a few research scholars are asking “Is that doing any good? Is that empowering them and giving them more knowledge and ability to execute responsibly ? “. Consider that, over the last three years, since the advent of generative AI tools, educational institutions have taken two distinct approaches. In one camp, they fully adopted without knowing their impact and following the general hype. In the other camp, which is shrinking as we speak, they failed to build fences around, without knowing what to fence.

The Ambition of Vision Language Models.

June 2025

Up until a few years ago nobody would expect computer vision systems to be able to talk. Up until then they would be buried in systems with limited capabilities and transparency. Recently,  we are experimenting with so-called Vision Language Models. Researchers’ ambition is to integrate the vision and language skills of AI systems towards a multimodal artificial intelligence.

 

 

2025 Funded PhD and RA Openings

At the Signals and Interactive Systems Lab (University of Trento, Italy) we are looking for highly motivated and talented graduate students to join our research team and work on Conversational Artificial Intelligence. This umbrella term includes the following research areas:

  • Natural Language Processing
  • Dialogue Modeling and Systems
  • Machine Learning
  • Affective Computing


We are investigating and designing next-generation ML models for multimodal input /output processing in physical and hybrid environments and interactions.

 

 

All news

Our Vision


We do research in the analysis and interpretation of signals generated and interpreted in human dialogue and interaction.

We design and train machine learning models to interpret these signals and support the inference process of human-machine dialogue and interaction systems. The design of the human-dialogue system is motivated by the goal of benefiting humans and evaluation methodologies and processes are a high priority in our lab. Our research and achievements come from collaborating with fantastic international researchers, professionals, non-profit organizations, and companies of all sizes.






Our Team

Open Thesis and Internships

Can Machines Really Align with Us? A Critical Evaluation of RLHF

Reinforcement Learning from Human Feedback (RLHF), in all its variations, has become the dominant method for aligning large language models with human preferences. In this thesis/internship, you will critically examine the effectiveness and limitations of RLHF. How robust is alignment across diverse users and contexts?  You’ll explore reward modeling, preference collection, and alignment drift to uncover whether current RLHF approaches truly capture human behavior—or it is a distant surrogate. Best suited for students interested in the foundations of machine-learning based alignment, ethics in AI, and empirical evaluation of learning systems.

Are Images Worth a Million Words ? a Million Names ? a Million Verbs ?

 

A long-standing research question aims to discover the amount of information in an image. At AT&T Bell Laboratories, a colleague claimed that images contained a lot of information, "a million words!". We argued about it, but it was too early back then in the '90s, and we left the topic in the Bell Labs blue notebook, the personal registry of each researcher. Now, more than ever, that category of research questions has surfaced. Many research studies investigate the relationship between visual and language modalities when building AI systems. Whether you are interested in research or applications or building systems, we are opening exciting opportunities in this area of research and applications.

Theory of Mind for LLMs ?

Theory of mind (ToM) is very much needed in humans to engage in social activity. Theory of mind is the ability to perceive and track the mental state of others. An example of ToM is the ability of humans to be empathic, that is the ability to reach the so-called self-other overlap state ( aka “put yourself in the other shoes” but it is more complex than that ). During the thesis/internship, you will assess the critical aspects of ToM in machines and neural models.

Generative AI vs. Information Retrieval

Join our cutting-edge research to fine-tune and evaluate state-of-the-art large language models (LLMs) for personalized response generation, and compare their performance against traditional response selection models. Gain hands-on experience, and contribute to groundbreaking advancements in Multimodal Human-Machine Dialogue and Interaction. Ideal for students and interns passionate about AI and machine learning!

Are LLMs Knowledgeable?

If LLMs answer factual questions, does it mean they are knowledgeable? Join our research on evaluating the uncertainty and accuracy of state-of-the-art LLMs in retrieving factual knowledge from their parameters. By analyzing their learned information during training, the study aims to enhance the understanding and reliability of LLMs in real-world applications. Ideal for students passionate about uncertainty and model evaluation.

Fine-Tuning vs. Retrieval-Augmented Generation: How Do Models Acquire Knowledge?

When large language models need to “know” something, should they memorize it—or look it up? This thesis/internship explores two competing strategies for knowledge acquisition: fine-tuning, which embeds information directly into model parameters, and Retrieval-Augmented Generation (RAG), which accesses external sources at inference time. You will evaluate how each approach supports factual accuracy, adaptability, and long-term retention. What does it mean for a model to know something? This project is ideal for students interested in the intersection of memory, reasoning, and knowledge representation in LLMs.

Publications

Filter:
All
Journals
Conferences
Books

  • Mousavi. M. S., Alghisi S. and Riccardi G., “ DyKnow: Dynamically Verifying Time-Sensitive Factual Knowledge in LLMs  ” Proc. Empirical Methods in Natural Language Processing (EMNLP), Miami, 2024.

  • Roccabruna G., Rizzoli M. and Riccardi G., “ Will LLMs Replace the Encoder-Only Models in Temporal Relation Classification?  ” Proc. Empirical Methods in Natural Language Processing (EMNLP), Miami, 2024.

  • Alghisi S., Rizzoli M., Roccabruna G., Mousavi. M. S.and Riccardi G., “ Should We Fine-Tune or RAG? Evaluating Different Techniques to Adapt LLMs for Dialogue  ” Proc. International Conference on Natural Langauge Generation, Tokyo, 2024.

  • Giuliano Tortoreto and Seyed Mahed Mousavi. 2024. "Dolomites@#SMM4H 2024: Helping LLMs “Know The Drill” in Low-Resource Settings - A Study on Social Media Posts." In Proceedings of The 9th Social Media Mining for Health Research and Applications (SMM4H 2024) Workshop and Shared Tasks, pages 17–22, Bangkok, Thailand. Association for Computational Linguistics.

  • M. Ravanelli, T. Parcollet, A. Moumen, S. de Langen, C. Subakan, P. Plantinga, Y. Wang, P. Mousavi, L. Della Libera, A. Ploujnikov, F. Paissan, D. Borra, S. Zaiem, Z. Zhao, S. Zhang, G. Karakasidis, S.-L. Yeh, P. Champion, A. Rouhe, R. Braun, F. Mai, J. Zuluaga-Gomez, S. M. Mousavi, et al., “Open-source conversational ai with speechbrain 1.0” Journal of Machine Learning Research, vol. 26, 2024.

  • Q. Chen, S. M. Mousavi, G. Jacucci, and G. Riccardi, “Investigating the use and perception of blocking feature in social virtual reality spaces,” Proceedings of the ACM on Human-Computer Interaction, vol. 9, 2024.

  • Mousavi. M. S., Roccabruna G., Alghisi S., Rizzoli M., Ravanelli M. and Riccardi G., " Are LLMs Robust for Spoken Dialogues? " Proc. International Workshop on Spoken Dialogue Systems Technology, TALK Sapporo, 2024.
  • Mousavi. M. S., Tanaka, S., Roccabruna G., Yoshino K., Nakamura S. and Riccardi G., " What’s New? Identifying the Unfolding of New Events in a Narrative " ACL, Proc. 5th Workshop of Narrative Understanding, 2023.
  • Mousavi. M. S., Caldarella S. and Riccardi G., " Response Generation in Longitudinal Dialogues: Which Knowledge Representation Helps? " ACL, Proc. 5th Workshop on NLP for Conversational AI, 2023.
  • Roccabruna G., Mousavi. M. S. and Riccardi G., " Understanding Emotion Valence is a Joint Deep Learning Task" ACL, Proc. 13th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis 2023.
  • Yin M., Roccabruna G., Azad A. and Riccardi G., " Let's Give a Voice to Conversational Agents in Virtual Reality " Proc. INTERSPEECH, Demo Paper , VIDEO Dublin, 2023.
  • Mayor-Torres J., Medina-DeVillers S., Clarkson T., Lerner M.D., Riccardi G. , "Evaluation of interpretability for deep learning algorithms in EEG emotion recognition: a case study in autism ", Artificial Intelligence in Medicine, Volume 143, 2023 .
  • Mousavi M., Roccabruna, Lorandi M., Caldarella S. and Riccardi G., " Evaluation of Response Generation Models: Shouldn’t It Be Shareable and Replicable? " EMNLP Workshop Generation, Evaluation & Metrics (GEM), 2022.
  • Bayerl S., Roccabruna G., Chowdhury A. S., Ciulli T., Danieli M., Riedhammer K. and Riccardi G., " What can Speech and Language Tell us About the Working Alliance in Psychotherapy " Proc. INTERSPEECH, 2022.
  • Mousavi M., Roccabruna G., Tammewar A. Azzolin S. and Riccardi G., " Can Emotion Carriers Explain Automatic Sentiment Prediction? A Study on Personal Narratives, " ACL, 12th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis, 2022.
  • Roccabruna G., Azzolin S. and Riccardi G., " Multi-source Multi-domain Sentiment Analysis with BERT-based Models, " Language Resources and Evaluation Conference, 2022.
  • Tammewar A., Bayerl P. S., Braun F., Riedhammer K. and Riccardi G., " Annotation of Valence for Spoken Personal Narratives, " Language Resources and Evaluation Conference, 2022.
  • Mousavi M., Negro R. and Riccardi G., " An Unsupervised Approach to Extract Life-Events from Personal Narratives in the Mental Health Domain, " Eighth Italian Conference on Computational Linguistics, 2022.
  • Danieli M., Ciulli T., Mousavi M. Silvestri G., Barbato S. Di Natale L. and Riccardi G., "Assessing the Impact of Conversational AI in the Treatment of Stress and Anxiety in Ageing Persons: A Randomized Controlled Trial Study ", Journal of Medical Internet Research (JMIR) Mental Health, Vol 9, No 9, 2022 .
  • Mayor-Torres J., Clarkson T., Hauschild K., Luhmann C. C., Lerner D. M. and Riccardi G., "Facial emotions are accurately encoded in the brains of those with autism: A deep learning approach", Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, Volume 7, Issue 7, Pages 688-695 2022 .
  • Bayerl P. S., Tammewar A., Riedhammer K. and Riccardi G., " Detecting Emotion Carriers By Combining Acoustic and Lexical Representations, " IEEE Automatic Speech Recognition and Understanding Conference, 2021.