MIOT : Towards Meta-Pruning via Optimal Transport

3 December 2024
from 17H00 to 18H00
Register
  • EN
  • Accessible by videoconference
  • Public
  • Guillaume EYNARD-BONTEMPS Guillaume EYNARD-BONTEMPS Ingénieur Calcul Scientifique

Pour le 5ème séminaire scientifique "Mathématiques & Informatique pour l'observation de la Terre" de la COMET TSI, nous avons le plaisir de recevoir Olin Geimer et Alexander Theus, doctorants à l'Université de Yale et à l'ETH Zurich, respectivement, le mardi 3 décembre de 17h00 à 18h00 en visioconférence.

Titre : Towards Meta-Pruning via Optimal Transport

Abstract :  Structural pruning of neural networks conventionally relies on identifying and  discarding less important neurons, a practice often resulting in significant accuracy loss that necessitates subsequent fine-tuning efforts. This paper introduces a novel approach named Intra-Fusion, challenging this prevailing pruning paradigm. Unlike existing methods that focus on designing meaningful neuron importance metrics, Intra-Fusion redefines the overlying pruning procedure. Through utilizing the concepts of model fusion and optimal transport, we leverage an agnostically given importance metric to arrive at a more  effective sparse model representation. Notably, our approach achieves substantial accuracy recovery without the need for resource-intensive fine-tuning, making it an efficient and promising tool for neural network compression.

Lien de la visio-conférence : https://rendez-vous.renater.fr/COMETTSIMIOT_e49abc-a09b19-ba5059

En espérant vous voir nombreux, n'oubliez pas de bloquer le créneau dans vos agendas !

Register
Add to your diary 3-12-2024 17:00 3-12-2024 18:00 <p>Pour le 5ème séminaire scientifique "Mathématiques &amp; Informatique pour l'observation de la Terre" de la COMET TSI, nous avons le plaisir de recevoir <a href="https://www.linkedin.com/in/olin-geimer/">Olin Geimer</a> et <a href="https://www.linkedin.com/in/alexander-theus-078b0a232/">Alexander Theus</a>, doctorants à l'Université de Yale et à l'ETH Zurich, respectivement, le <strong>mardi 3 décembre de 17h00 à 18h00 </strong>en visioconférence.</p> <p>Titre : <strong>Towards Meta-Pruning via Optimal Transport</strong><br /> <br /> Abstract :&nbsp; <em>Structural pruning of neural networks conventionally relies on identifying and&nbsp; discarding less important neurons, a practice often resulting in significant accuracy loss that necessitates subsequent fine-tuning efforts. This paper introduces a novel approach named Intra-Fusion, challenging this prevailing pruning paradigm. Unlike existing methods that focus on designing meaningful neuron importance metrics, Intra-Fusion redefines the overlying pruning procedure. Through utilizing the concepts of model fusion and optimal transport, we leverage an agnostically given importance metric to arrive at a more&nbsp; effective sparse model representation. Notably, our approach achieves substantial accuracy recovery without the need for resource-intensive fine-tuning, making it an efficient and promising tool for neural network compression.</em></p> <p>Lien de la visio-conférence :&nbsp;<a href="https://rendez-vous.renater.fr/COMETTSIMIOT_e49abc-a09b19-ba5059">https://rendez-vous.renater.fr/COMETTSIMIOT_e49abc-a09b19-ba5059</a></p> <p>En espérant vous voir nombreux, n'oubliez pas de bloquer le créneau dans vos agendas !</p>

To take part in this event, you must register.

Share :

Partage possible par tous (via Facebook, LinkedIn, Twitter, email)

You may find these events interesting, as well