Francesco Paissan
 |
Francesco Paissan, |
I am a student at the University of Trento, currently enrolled in the computer engineering undergrad program. My research focuses on designing efficient neural networks and training paradigms that enable fast, low-footprint inference and on-device network training. Following my research internship at Mila, I am collaborating with Prof. Subakan and Prof. Ravanelli on interpretability techniques for neural networks, specifically for speech and audio. I am a contributor of the SpeechBrain project, and one of the core contributors of the SpeechBrain-MOABB benchmark. I co-organized the ICASSP 2024 XAI-SA workshop.
You can see my google scholar here.
Research Interests
Resource-efficient machine/deep learning
Interpretable Machine Learning, Explainability, Posthoc Interpretations
Audio generation/editing using neural models
Machine Learning, Statistical Learning
Representative papers
Listenable Maps for Audio Classifiers [demo] [slides] [video] [poster]
tinyCLAP: Distilling Constrastive Language-Audio Pretrained Models [demo] [slides]
Talks, seminars, lectures
tinyML EMEA Innovation Forum - "tinyCLAP: towards embedded foundational models" [slides]
UniPD Elements of Deep Learning doctoral course - "An introduction to tinyML" [slides]
tinyML Foundation webcast series - "tinyML: Designing Ecient Neural Archictures and Scaling Strategies for Edge Computing" [slides]
UniPD DEI Seminar - "tinyML: neural networks design principles, scaling strategies and beyond" [slides]
IEEE World Forum on IoT - "Hands‐on tinyML for IoT, bringing intelligence to the edge" [slides]
INFN/UniRoma3 Seminar - "Emerging Opportunities of Machine Learning in Physics"
Timeline
We have two papers accepted at ICASSP 2025! One in the main conference, and one in the SPADE workshop!
Our Listenable Maps for Zero-Shot Audio Classifiers was accepted at NeurIPS 2024.
Our Listenable Maps for Audio Classifiers was accepted at ICML 2024 as Oral!
Our tinyCLAP and Audio Editing with Non-rigid Text Prompt papers were accepted to Interspeech 2024.
Many things, I decided to start doing this too late. :)
Contact
email: francescopaissan at gmail dot com
twitter: @fpaissan_
bluesky: @fpaissan.bsky.social