Autoattention multitêtes - Historique des versions

Pitpitt : Remplacement de texte : « == Anglais == » par «  == Anglais == »

2026-07-29T02:36:24Z

Remplacement de texte : « == Anglais == » par « <span id="en"></span> == Anglais == »

← Version précédente		Version du 28 juillet 2026 à 22:36
Ligne 17 :		Ligne 17 :
	'''attention multi-têtes'''		'''attention multi-têtes'''

			<span id="en"></span>
	== Anglais ==		== Anglais ==
	''' multi-head self-attention'''		''' multi-head self-attention'''

Pitpitt : Remplacement de texte : « ==Español== » par «  ==Español== »

2026-07-29T00:11:43Z

Remplacement de texte : « ==Español== » par « <span id="es"></span> ==Español== »

← Version précédente		Version du 28 juillet 2026 à 20:11
Ligne 23 :		Ligne 23 :
	In deep learning, multi-head self-attention is the processing of a word string (typically in natural language) by parallelizing the attention mechanism at several points in the sequence and then merging the results.		In deep learning, multi-head self-attention is the processing of a word string (typically in natural language) by parallelizing the attention mechanism at several points in the sequence and then merging the results.

			<span id="es"></span>
	==Español==		==Español==
	[[Catégorie:es]]		[[Catégorie:es]]

Pitpitt le 18 avril 2026 à 15:36

2026-04-18T15:36:50Z

← Version précédente		Version du 18 avril 2026 à 11:36
Ligne 21 :		Ligne 21 :

	''' multi-head attention'''		''' multi-head attention'''
	<!-- ~~Multi-head attention is~~ a ~~module for attention mechanisms which runs through an~~ attention mechanism several ~~times~~ in ~~parallel. The independent attention outputs are~~ then ~~concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of~~ the ~~sequence differently (e.g. longer-term dependencies versus shorter-term dependencies)~~. ~~-->~~		In deep learning, multi-head self-attention is the processing of a word string (typically in natural language) by parallelizing the attention mechanism at several points in the sequence and then merging the results.

	==Español==		==Español==

Pitpitt : Remplacement de texte : « ==Español== » par « ==Español== Catégorie:es »

2025-09-24T00:57:45Z

Remplacement de texte : « ==Español== » par « ==Español== Catégorie:es »

← Version précédente		Version du 23 septembre 2025 à 20:57
Ligne 24 :		Ligne 24 :

	==Español==		==Español==
			[[Catégorie:es]]

	''''' autoatención multicéfala '''''		''''' autoatención multicéfala '''''

Jean-Sébastien Zavalone le 21 juillet 2025 à 19:34

2025-07-21T19:34:35Z

← Version précédente		Version du 21 juillet 2025 à 15:34
Ligne 22 :		Ligne 22 :
	''' multi-head attention'''		''' multi-head attention'''
	<!-- Multi-head attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->		<!-- Multi-head attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->

			==Español==

			''''' autoatención multicéfala '''''

			''En el aprendizaje profundo, la autoatención multicabezal es el procesamiento de una secuencia (normalmente en lengua natural) paralelizando el mecanismo de autoatención en varios puntos de la secuencia y fusionando después los resultados.''

	== Source ==		== Source ==
	[https://infoscience.epfl.ch/record/300271/files/EPFL_TH9822.pdf Cordonnier, J.-B. (2023), ''Transformer Models for Vision''.]		[https://infoscience.epfl.ch/record/300271/files/EPFL_TH9822.pdf Cordonnier, J.-B. (2023), ''Transformer Models for Vision''.]
Ligne 28 :		Ligne 35 :

	[https://cs.paperswithcode.com/method/multi-head-attention Vaswani et al. (2017) ''Attention Is All You Need'']		[https://cs.paperswithcode.com/method/multi-head-attention Vaswani et al. (2017) ''Attention Is All You Need'']
			{{Modèle:101}}

	[[Catégorie:GRAND LEXIQUE FRANÇAIS]]		[[Catégorie:GRAND LEXIQUE FRANÇAIS]]

	[[Catégorie:101]]		[[Catégorie:101]]

Arianne le 8 février 2025 à 18:42

2025-02-08T18:42:32Z

← Version précédente		Version du 8 février 2025 à 14:42
Ligne 20 :		Ligne 20 :
	''' multi-head self-attention'''		''' multi-head self-attention'''

	''' ~~Mmulti~~-head attention'''		''' multi-head attention'''
	<!-- Multi-head attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->		<!-- Multi-head attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->
	== Source ==		== Source ==

Pitpitt le 16 janvier 2025 à 01:16

2025-01-16T01:16:45Z

← Version précédente		Version du 15 janvier 2025 à 21:16
Ligne 30 :		Ligne 30 :

	[[Catégorie:GRAND LEXIQUE FRANÇAIS]]		[[Catégorie:GRAND LEXIQUE FRANÇAIS]]

			[[Catégorie:101]]

Arianne le 12 janvier 2025 à 00:31

2025-01-12T00:31:33Z

← Version précédente		Version du 11 janvier 2025 à 20:31
Ligne 1 :		Ligne 1 :
	== Définition ==		== Définition ==

	En [[apprentissage profond]], l'~~[[Mécanisme d'attention\|~~autoattention]] multitêtes est un traitement de [[Séquence de mots\|séquence]] (typiquement en [[langue naturelle]]) qui consiste à paralléliser le mécanisme d'[[Mécanisme d'attention\|autoattention]] en plusieurs points de la ~~[[Séquence de mots\|~~séquence]] pour ensuite fusionner les résultats.		En [[apprentissage profond]], l'autoattention multitêtes est un traitement de [[Séquence de mots\|séquence]] (typiquement en [[langue naturelle]]) qui consiste à paralléliser le mécanisme d'[[Mécanisme d'attention\|autoattention]] en plusieurs points de la séquence pour ensuite fusionner les résultats.

	== Compléments ==		== Compléments ==
Ligne 9 :		Ligne 9 :

	== Français ==		== Français ==
	''' ~~Autoattention~~ multitêtes '''		''' autoattention multitêtes '''

	''' ~~Autoattention~~ multi-têtes '''		''' autoattention multi-têtes '''

	'''~~Attention~~ multitêtes'''		'''attention multitêtes'''

	'''~~Attention~~ multi-têtes'''		'''attention multi-têtes'''

	== Anglais ==		== Anglais ==
	''' ~~Multi~~-~~Head Attention~~'''		''' multi-head self-attention'''

	''' ~~Multi~~-~~Head Self-Attention~~'''		''' Mmulti-head attention'''
	<!-- Multi-head ~~Attention is~~ a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->		<!-- Multi-head attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->
	== Source ==		== Source ==
	[https://infoscience.epfl.ch/record/300271/files/EPFL_TH9822.pdf Cordonnier, J.-B. (2023), ''Transformer Models for Vision''.]		[https://infoscience.epfl.ch/record/300271/files/EPFL_TH9822.pdf Cordonnier, J.-B. (2023), ''Transformer Models for Vision''.]

Pitpitt le 4 septembre 2024 à 23:54

2024-09-04T23:54:59Z

← Version précédente		Version du 4 septembre 2024 à 19:54
Ligne 29 :		Ligne 29 :
	[https://cs.paperswithcode.com/method/multi-head-attention Vaswani et al. (2017) ''Attention Is All You Need'']		[https://cs.paperswithcode.com/method/multi-head-attention Vaswani et al. (2017) ''Attention Is All You Need'']

	[[Catégorie:~~Publication~~]]		[[Catégorie:GRAND LEXIQUE FRANÇAIS]]

Pitpitt le 4 septembre 2024 à 23:53

2024-09-04T23:53:16Z

← Version précédente		Version du 4 septembre 2024 à 19:53
Ligne 21 :		Ligne 21 :

	''' Multi-Head Self-Attention'''		''' Multi-Head Self-Attention'''
			<!-- Multi-head Attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies). -->
	<!-- Multi-head Attention is a module for attention mechanisms which runs through an attention mechanism several times in parallel. The independent attention outputs are then concatenated and linearly transformed into the expected dimension. Intuitively, multiple attention heads allows for attending to parts of the sequence differently (e.g. longer-term dependencies versus shorter-term dependencies).
	-->

	== Source ==		== Source ==
	[https://infoscience.epfl.ch/record/300271/files/EPFL_TH9822.pdf Cordonnier, J.-B. (2023), ''Transformer Models for Vision''.]		[https://infoscience.epfl.ch/record/300271/files/EPFL_TH9822.pdf Cordonnier, J.-B. (2023), ''Transformer Models for Vision''.]