English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Prototype-based meta-learning has emerged as a powerful technique foraddressing few-shot learning challenges. However, estimating a deterministicprototype using a simple average function from a limited number of examplesremains a fragile process. To overcome this limitation, we introduce ProtoDiff,a novel framework that leverages a task-guided diffusion model during themeta-training phase to gradually generate prototypes, thereby providingefficient class representations. Specifically, a set of prototypes is optimizedto achieve per-task prototype overfitting, enabling accurately obtaining theoverfitted prototypes for individual tasks. Furthermore, we introduce atask-guided diffusion process within the prototype space, enabling themeta-learning of a generative process that transitions from a vanilla prototypeto an overfitted prototype. ProtoDiff gradually generates task-specificprototypes from random noise during the meta-test stage, conditioned on thelimited samples available for the new task. Furthermore, to expedite trainingand enhance ProtoDiff's performance, we propose the utilization of residualprototype learning, which leverages the sparsity of the residual prototype. Weconduct thorough ablation studies to demonstrate its ability to accuratelycapture the underlying prototype distribution and enhance generalization. Thenew state-of-the-art performance on within-domain, cross-domain, and few-taskfew-shot classification further substantiates the benefit of ProtoDiff.

ProtoDiff: Learning to Learn Prototypical Networks by Task-Guided Diffusion