Davide Buffelli

Ph.D. · AI Researcher · davide.buffelli [at] mtkresearch.com

I am a Senior Deep Learning Researcher at MediaTek Research in London (UK). I have a broad interest in Deep Learning, with a current focus on foundation models for multimodal time-series, optimization, and representation learning for structured data.

I hold a Ph.D. in Information Engineering from the University of Padova, where I focused on Graph Neural Networks and Graph Representation Learning and I was supervised by Professor Fabio Vandin. During my PhD I have been a Research Scientist Intern at Meta AI (London) and at Samsung AI Research (Cambridge), and a visiting student in Professor Pietro Liò's group at the University of Cambridge, and in Dr. Bastian Rieck's group at Helmholtz Munich. You check out my full CV here.


News

11/2024 Our paper "CliquePH: Higher-Order Information for Graph Neural Networks through Persistent Homology on Clique Graphs" has been accepted at LoG 2024! Great collaboration with Farzin Soleymani and Bastian Rieck.
09/2024 Two papers accepted at NeurIPS 2024! "Exact, Tractable Gauss-Newton Optimization in Deep Reversible Architectures Reveal Poor Generalization" with colleagues from MediaTek Research, and "Deep Equilibrium Algorithmic Reasoning" in collaboration with Dobrik Georgiev, JJ Wilson, and Pietro Liò .
04/2024 Our paper "The Deep Equilibrium Algorithmic Reasoner" has been accepted at the CVPR Workshop on Multimodal Algorithmic Reasoning! This is a collaboration with the amazing Dobrik Georgiev and Pietro Liò on a new model for Neural Algorithmic Reasoning.
05/2023 I am starting a new position as a Senior Deep Learning Researcher at MediaTek Research! I look forward to working with this amazing team.
03/2023 I have successfuly defended my PhD thesis (available here)! I look forward to the next steps in my career.
11/2022 Our paper "Scalable Theory-Driven Regularization of Scene Graph Generation Models" has been accepted for publication at AAAI 2023!
09/2022 Our paper "SizeShiftReg: a Regularization Method for Improving Size-Generalization in Graph Neural Networks" has been accepted for publication at NeurIPS 2022!
09/2022 I am starting as a Research Scientist Intern at Meta AI in London!

View Older News


Publications


  • Exact, Tractable Gauss-Newton Optimization in Deep Reversible Architectures Reveal Poor Generalization
    Davide Buffelli*, Jamie McGowan*, Wangkun Xu, Alexandru Cioba, Da-shan Shiu, Guillaume Hennequin, Alberto Bernacchia, Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS), 2024.
    [Paper coming soon] [PDF (arXiv)]
  • Deep Equilibrium Algorithmic Reasoning
    Dobrik Georgiev, JJ Wilson, Davide Buffelli, Pietro Liò, Thirty-Eighth Annual Conference on Neural Information Processing Systems (NeurIPS), 2024.
    [Paper coming soon] [PDF (arXiv)]
  • CliquePH: Higher-Order Information for Graph Neural Networks through Persistent Homology on Clique Graphs
    Davide Buffelli*, Farzin Soleymani*, Bastian Rieck, Learning on Graphs (LoG), 2024.
    [Paper coming soon] [PDF (arXiv)]
  • The Deep Equilibrium Algorithmic Reasoner
    Dobrik Georgiev, Pietro Liò, Davide Buffelli, CVPR Workshop on Multimodal Algorithmic Reasoning, 2024. (Spotlight)
    [Paper] [PDF (arXiv)]
  • Is Meta-Learning the Right Approach for the Cold-Start Problem in Recommender Systems?
    Davide Buffelli, Ashish Gupta, Agnieszka Strzalka, Vassilis Plachouras, Preprint, 2023.
    [PDF (arXiv)]
  • Improving the Effectiveness of Graph Neural Networks in Practical Scenarios
    Davide Buffelli, PhD Thesis, University of Padova, 2023.
    [Full Text]
  • Scalable Theory-Driven Regularization of Scene Graph Generation Models
    Davide Buffelli*, Efthymia Tsamoura*, Thirty-Seventh AAAI Conference on Artificial Intelligence (AAAI), 2023.
    [Paper] [PDF (arXiv)]
  • SizeShiftReg: a Regularization Method for Improving Size-Generalization in Graph Neural Networks
    Davide Buffelli, Pietro Liò, Fabio Vandin, Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS), 2022.
    [Paper] [PDF (arXiv)] [Code]
  • Graph Representation Learning for Multi-Task Settings: a Meta-Learning Approach
    Davide Buffelli, Fabio Vandin, International Joint Conference on Neural Networks (IJCNN), 2022. (Oral)
    [Paper] [PDF (arXiv)] [Code]
  • The Impact of Global Structural Information in Graph Neural Networks Applications
    Davide Buffelli, Fabio Vandin, Data (special issue "Knowledge Extraction from Data Using Machine Learning"), 2022.
    [Paper] [PDF (arXiv)] [Code]
  • Extending Logic Explained Networks to Text Classification
    Rishabh Jain, Gabriele Ciravegna, Pietro Barbiero, Francesco Giannini, Davide Buffelli, Pietro Liò, The 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2022.
    [Paper] [PDF (arXiv)]
  • Attention-Based Deep Learning Framework for Human Activity Recognition with User Adaptation
    Davide Buffelli, Fabio Vandin, IEEE Sensors Journal, 2021.
    [Paper] [PDF (arXiv)] [Code]
  • A Meta-Learning Approach for Graph Representation Learning in Multi-Task Settings
    Davide Buffelli, Fabio Vandin, NeurIPS Workshop on Meta-Learning (MetaLearn), 2020.
    [PDF] [PDF (arXiv; with Appendix)] [Video] [Slides] [Poster] [Code]

You can send me an email at davide.buffelli [at] mtkresearch.com, and you can find me on Google Scholar, ORCID, LinkedIn, GitHub, and Twitter.