Overview

My research interests

  • Contents: my research activity is placed in the wide discipline of statistical-mechanics methods for complex systems analysis, in particular concerning spin-glass theory and its application to modern Artificial Intelligence and Machine Learning. My research is focused in particular on attractor neural networks (ANNs) as prototypical model for associative memory, such as the Hopfield model and its related variants. Also, I investigate collective behaviors of archetypical generative models, such as the Restricted Boltzmann Machines, with the aim of developing an interpretable spin-glass theory depicting their working regimes.
  • Methods: a crucial point of my research is the development of rigorous techniques towards a solid-grounded mathematical theory of Artificial Intelligence and Machine Learning. In particular, the technology adopted falls in the fields of Guerra's interpolation schemes, PDE theory (in particular, Hamilton-Jacobi), probability and random matrix theory.
  • Other interests: I also dealt with statistical inference for biological problems, in particular concerning non-linear analysis for heart rate variability and in vitro diffusion experiments quantifying the effects of chemioterapic drugs on mutual stroma-cancer cells interaction for different pancreatic lines with a composite approach of maximum entropy principle, graph theory and (multi-layer feed-forward) neural networks design.

Keywords

Neural Networks; Machine Learning; Deep Learning; Artificial Intelligence; Spin-glass Theory; Statistical Physics; Mathematical Physics; Theoretical Physics.

Selected papers

  • Regularization, early-stopping and dreaming: a Hopfield-like setup to address generalization and overfitting
    E. Agliari, F. Alemanno, M. Aquaro, A. Fachechi
    Neural Networks 177, 106389 (2024)
    Total number of citations: 5
  • Quantifying heterogeneity to drug response in cancer–stroma kinetics
    F. Alemanno, M. Cavo, D. Delle Cave, A. Fachechi, R. Rizzo, E. D’Amone, G. Gigli, E. Lonardo, A. Barra, L. L Del Mercato
    Proceedings of the National Academy of Sciences 120 (11), e2122352120 (2023)
    Total number of citations: 3
  • Outperforming RBM feature-extraction capabilities by “dreaming” mechanism
    A. Fachechi, A. Barra, E. Agliari, F. Alemanno
    IEEE transactions on neural networks and learning systems 35 (1), 1172-1181 (2022)
    Total number of citations: 16
  • Neural networks with a redundant representation: Detecting the undetectable
    E. Agliari, F. Alemanno, A. Barra, M. Centonze, A. Fachechi
    Physical review letters 124 (2), 028301 (2020)
    Total number of citations: 36
  • Dreaming neural networks: forgetting spurious memories and reinforcing pure ones
    A. Fachechi, E. Agliari, A Barra
    Neural Networks 112, 24-40 (2019)
    Total number of citations: 62
  • Virasoro vacuum block at next-to-leading order in the heavy-light limit
    M. Beccaria, A. Fachechi, G. Macorini
    Journal of High Energy Physics 2016 (2), 1-22 (2016)
    Total number of citations: 53

Additional informations

  • Since December 2024, I am member of the Department Committee for informatic resources.
  • On November 2024, I earned the National Certification for Associate Professor in Mathematical Physics (Abilitazione Scientifica Nazionale per Professore di II fascia, SC: 01/A4), see the final judgement of the MUR committee (in italian).
  • Since April 2023, I have been part of the FAIR foundation, a PNRR-funded project aiming at high-quality research in the field of Artificial Intelligence. Specifically, I am a Researcher for the Spoke 5 (WP5.5) with main research line "Quality assessment in Hard Sciences and AI". I am also responsible for communication for the organization of dissemination events within the FAIR group.
  • My recent research activity has benefited from the stimulating environment provided by the Alan Turing Institute’s event “Physics-informed Machine Learning”, which took place in London from 16 to 23 January, 2023.
  • My work on "Dreaming Neural Networks" has attracted much attention from the general press, see for example this article.
  • Since 2018, I have been member of the National Group of Mathematical Physics (GNFM-INdAM), within the community of Mechanics for Discrete Systems.