4 - Seminar Meta Learning (SemMeL) - Jonas Utz - Prototypical Networks for Few-shot Learning/ClipID:24667 previous clip next clip

Keywords: meta learning
The automatic subtitles generated using Whisper Open AI in this video player (and in the Multistream video player) are provided for convenience and accessibility purposes. However, please note that accuracy and interpretation may vary. For more information, please refer to the FAQs (Paragraph 14).
Recording date 2020-11-23

Via

Free

Language

English

Organisational Unit

Friedrich-Alexander-Universität Erlangen-Nürnberg

Producer

Friedrich-Alexander-Universität Erlangen-Nürnberg

Today Jonas Utz presents the paper "Prototypical Networks for Few-shot Learning"

We propose Prototypical Networks for the problem of few-shot classification, where a classifier must generalize to new classes not seen in the training set, given only a small number of examples of each new class. Prototypical Networks learn a metric space in which classification can be performed by computing distances to prototype representations of each class. Compared to recent approaches for few-shot learning, they reflect a simpler inductive bias that is beneficial in this limited-data regime, and achieve excellent results. We provide an analysis showing that some simple design decisions can yield substantial improvements over recent approaches involving complicated architectural choices and meta-learning. We further extend Prototypical Networks to zero-shot learning and achieve state-of-the-art results on the CU-Birds dataset.

https://arxiv.org/abs/1703.05175

More clips in this category "Friedrich-Alexander-Universität Erlangen-Nürnberg"

2024-02-12
IdM-login / Passwort / Studon
protected  
2024-02-07
IdM-login
protected  
2024-02-06
IdM-login / Studon
protected  
2024-02-06
Studon
protected