publications

publications by categories in reversed chronological order.

2026

  1. ICLR
    paper_injective.png
    Language Models are Injective and Hence Invertible
    Giorgos Nikolaou*, Tommaso Mencattini*, Donato Crisostomi, and 3 more authors
    In ICLR, 2026
    Paper
  2. Oral
    ICLR (Oral)
    saence.png
    Exploratory Causal Inference in SAEnce
    Tommaso Mencattini*, Riccardo Cadei*, and Francesco Locatello
    In ICLR (Oral), 2026
    Paper

2025

  1. ACL (Demo)
    mergenetic.png
    Mergenetic: a Simple Evolutionary Model Merging Library
    Adrian Robert Minut*, Tommaso Mencattini*, Andrea Santilli, and 2 more authors
    In ACL (Demo), 2025
    Paper
  2. activation_patching.png
    Activation Patching for Interpretable Steering in Music Generation
    Simone Facchiano, Giorgio Strano, Donato Crisostomi, and 4 more authors
    2025
    Paper
  3. ICML
    merge3.png
    MERGE^3: Efficient Evolutionary Merging on Consumer-grade GPUs
    Tommaso Mencattini*, Adrian Robert Minut*, Donato Crisostomi, and 2 more authors
    In ICML, 2025
    Paper
  4. UniReps Workshop @ NeurIPS 2024, BlogPost track
    sae.png
    Do Sparse Autoencoders (SAEs) transfer across base and finetuned language models?
    Taras Kutsyk*, Tommaso Mencattini*, and Ciprian Florea*
    UniReps Workshop @ NeurIPS 2024, BlogPost track, 2025
    Blog post
    Paper

2023

  1. Intelligent Information Systems
    cicero.png
    Cicero: An AI-Based Writing Assistant for Legal Users
    Francesca De Luzi, Mattia Macrì, Massimo Mecella, and 1 more author
    In Intelligent Information Systems, 2023
    Paper