Deep Learning Compression & Hugging Face Hub

Wednesday 02.08.2023
Time:
From 14:30 to 17:00

Address:

Amphitéatre du Pôle Numérique Brest Iroise (PNBI)

Public

Enseignants chercheurs
Doctorants
Etudiants
BRAIn team's seminar
Deep Learning Compression & Hugging Face Hub
Wednesday, February 8,
Amphitéatre du Pôle Numérique Brest Iroise (PNBI)
 
2:30 pm: Stefan Duffner
Reducing the complexity of Convolutional Neural Networks by quantization and filter pruning

 

Deep Neural Networks are powerful machine learning models that have obtained excellent performance for numerous applications. However, they may require a large amount of computational and memory resources leading to a high energy consumption making their use for embedded systems difficult. In this talk, I will present two different approaches that we proposed recently to reduce this complexity. This first is a post-training quantization method that approximates the filters of a CNN using dyadic rationals such that all multiplications of the model can be replaced by bit shifts and additions. The second approach is a compression-aware training algorithm, that performs filter pruning on a CNN by introducing a new sparsity-inducing regularization term based on the l1/l2 pseudo norm. Both approach have been applied to standard computer vision benchmarks and can considerably reduce the complexity of CNN with only a small precision loss.

 

3:50 pm: Eliott Coyac
A tour through the Hugging Face Hub
 
The Hugging Face Hub is a platform where data scientists can share their ML models, download largedatasets, and showcase their models for conferences through interactive demos with a few lines of Python code. With integrations to ArXiv, libraries to optimize models and services to deploy them in production, it helps bridge the gap between research and industry.

 

See more

L'équipe BRAIn vous convie à un séminaire le 8 février 2023 au cours duquel nous avons le plaisir d'accueillir Stefan Duffner, Maître de Conférences à l'INSA Lyon au laboratoire LIRIS, ainsi que Eliott Coyac de Hugging Face.

Stefan nous parlera de compression de réseaux de neurones, tandis que Eliott nous présentera un nouvel outil de partage de modèle et datasets de Machine Learning développé par Hugging Face.

Organizer(s)

Équipe de Recherche :  BRAIn (Better Representations for Artificial Intelligence)

Partner(s)

INSA Lyon

LIRIS

Published on 03.02.2023
 
1
 
2
 
3
 
4
 
5
 
6
 
7
 
8
 
9
 
10
 
11
 
12
 
13
 
14
 
15
 
16
 
17
 
18
 
19
 
20
 
21
 
22
 
23
 
24
 
25
 
26
 
27
 
28
 
29
 
30
 
31