UDON: Universal Dynamic Online distillatioN for generic image representations

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F24%3A00382301" target="_blank" >RIV/68407700:21230/24:00382301 - isvavai.cz</a>

Výsledek na webu

<a href="https://papers.nips.cc/paper_files/paper/2024/hash/9df6a759e01b871a009138bcc0471607-Abstract-Conference.html" target="_blank" >https://papers.nips.cc/paper_files/paper/2024/hash/9df6a759e01b871a009138bcc0471607-Abstract-Conference.html</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

UDON: Universal Dynamic Online distillatioN for generic image representations

Popis výsledku v původním jazyce

Universal image representations are critical in enabling real-world fine-grained and instance-level recognition applications, where objects and entities from any domain must be identified at large scale. Despite recent advances, existing methods fail to capture important domain-specific knowledge, while also ignoring differences in data distribution across different domains. This leads to a large performance gap between efficient universal solutions and expensive approaches utilising a collection of specialist models, one for each domain. In this work, we make significant strides towards closing this gap, by introducing a new learning technique, dubbed UDON (Universal Dynamic Online distillatioN). UDON employs multi-teacher distillation, where each teacher is specialized in one domain, to transfer detailed domain- specific knowledge into the student universal embedding. UDON’s distillation approach is not only effective, but also very efficient, by sharing most model parameters between the student and all teachers, where all models are jointly trained in an online manner. UDON also comprises a sampling technique which adapts the training process to dynamically allocate batches to domains which are learned slower and require more frequent processing. This boosts significantly the learning of complex domains which are characterised by a large number of classes and long- tail distributions. With comprehensive experiments, we validate each component of UDON, and showcase significant improvements over the state of the art in the recent UnED benchmark. Code: https://github.com/nikosips/UDON.

Název v anglickém jazyce

UDON: Universal Dynamic Online distillatioN for generic image representations

Popis výsledku anglicky

Universal image representations are critical in enabling real-world fine-grained and instance-level recognition applications, where objects and entities from any domain must be identified at large scale. Despite recent advances, existing methods fail to capture important domain-specific knowledge, while also ignoring differences in data distribution across different domains. This leads to a large performance gap between efficient universal solutions and expensive approaches utilising a collection of specialist models, one for each domain. In this work, we make significant strides towards closing this gap, by introducing a new learning technique, dubbed UDON (Universal Dynamic Online distillatioN). UDON employs multi-teacher distillation, where each teacher is specialized in one domain, to transfer detailed domain- specific knowledge into the student universal embedding. UDON’s distillation approach is not only effective, but also very efficient, by sharing most model parameters between the student and all teachers, where all models are jointly trained in an online manner. UDON also comprises a sampling technique which adapts the training process to dynamically allocate batches to domains which are learned slower and require more frequent processing. This boosts significantly the learning of complex domains which are characterised by a large number of classes and long- tail distributions. With comprehensive experiments, we validate each component of UDON, and showcase significant improvements over the state of the art in the recent UnED benchmark. Code: https://github.com/nikosips/UDON.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach<br>I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2024

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Proceedings of the NeurIPS

ISBN

979-8-3313-1438-5

ISSN

1049-5258

e-ISSN

—

Počet stran výsledku

24

Strana od-do

—

Název nakladatele

Neural Information Processing Systems (NIPS) Foundation

Místo vydání

—

Místo konání akce

Vancouver

Datum konání akce

10. 12. 2024

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—