Hello, I am Matteo Attimonelli, a PhD candidate in Artificial Intelligence in the National PhD Program at Sapienza University of Rome, hosted at the Information Systems Research Group (SisInfLab) of the Polytechnic University of Bari, lead by Prof. Tommaso Di Noia.

My research focuses on multimodal representation learning with LLMs and large vision–language models (LVLMs) for retrieval and recommendation: building generalist embeddings on top of them, auditing their benchmarks, and improving their robustness and efficiency (sparse vs. dense). My investigation started from multimodal recommendation and controllable image generation and evolved toward LVLM-centric retrieval, RAG, and personalization.

I have been a visiting PhD student at the University of Edinburgh (Dr. Pasquale Minervini’s group), auditing composed image retrieval benchmarks and studying the limits and robustness of embedding-based retrieval, and I am currently visiting the Alpen-Adria-Universität Klagenfurt (Prof. Dietmar Jannach’s group), exploring alternatives to classical BPR-style optimization for personalization. Before the PhD, I spent almost two years in industry as a Deep Learning Engineer, building computer vision and recommendation solutions for practical industrial problems.

Recent publications

Bold style means (one of) main author(s). Full list here.

• Do Composed Image Retrieval Benchmarks Require Multimodal Composition?
  Attimonelli Matteo, De Bellis Alessandro, Gema Aryo Pradipta, Saxena Rohit, Sekoyan Monica, Kwan Wai-Chung, Pomo Claudio, Suglia Alessandro, Jannach Dietmar, Di Noia Tommaso, Minervini Pasquale
  arXiv:2605.14787, under review, 2026.

• WaveDiT: Distribution-Aware Wavelet Flow Matching for Efficient 3D Brain MRI Synthesis
  Danese Danilo, Lombardi Angela, Fasano Giuseppe, Attimonelli Matteo, Di Noia Tommaso
  MICCAI, 2026.

• FlowLet: Conditional 3D Brain MRI Synthesis using Wavelet Flow Matching
  Danese Danilo, Lombardi Angela, Attimonelli Matteo, Fasano Giuseppe, Di Noia Tommaso
  Medical Image Analysis, 2026.

• GeCo: Towards Effective GAN-Based Fashion Compatibility Modeling and Retrieval
  Attimonelli Matteo, Pomo Claudio, Jannach Dietmar, Di Noia Tommaso
  Information Sciences, 2026.

• Large-Scale Benchmarks for Multimodal Recommendation with Ducho
  Attimonelli Matteo, Danese Danilo, Di Fazio Angela, Malitesta Daniele, Pomo Claudio, Di Noia Tommaso
  Expert Systems with Applications, 2025.

• Do Recommender Systems Really Leverage Multimodal Content? A Comprehensive Analysis on Multimodal Representations for Recommendation
  Pomo Claudio, Attimonelli Matteo, Danese Danilo, Narducci Fedelucio, Di Noia Tommaso
  In Proceedings of the 34th ACM International Conference on Information and Knowledge Management (CIKM ‘25), November 10–14, 2025, Seoul, Republic of Korea.

• Do We Really Need Specialization? Evaluating Generalist Text Embeddings for Zero-Shot Recommendation and Search
  Attimonelli Matteo, De Bellis Alessandro, Pomo Claudio, Jannach Dietmar, Di Sciascio Eugenio, Di Noia Tommaso
  In Proceedings of the Nineteenth ACM Conference on Recommender Systems (RecSys ‘25), September 22–26, 2025, Prague, Czech Republic.

Contacts

• institutional email: matteo.attimonelli@poliba.it
• personal email: matteo.attimonelli1999@gmail.com
• institutional webpage: https://sisinflab.poliba.it/people/matteo-attimonelli/