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

Foundation models have shown outstanding performance and generalizationcapabilities across domains. Since most studies on foundation models mainlyfocus on the pretraining phase, a naive strategy to minimize a singletask-specific loss is adopted for fine-tuning. However, such fine-tuningmethods do not fully leverage other losses that are potentially beneficial forthe target task. Therefore, we propose MEta Loss TRansformer (MELTR), a plug-inmodule that automatically and non-linearly combines various loss functions toaid learning the target task via auxiliary learning. We formulate the auxiliarylearning as a bi-level optimization problem and present an efficientoptimization algorithm based on Approximate Implicit Differentiation (AID). Forevaluation, we apply our framework to various video foundation models (UniVL,Violet and All-in-one), and show significant performance gain on all fourdownstream tasks: text-to-video retrieval, video question answering, videocaptioning, and multi-modal sentiment analysis. Our qualitative analysesdemonstrate that MELTR adequately `transforms' individual loss functions and`melts' them into an effective unified loss. Code is available athttps://github.com/mlvlab/MELTR.

MELTR: Meta Loss Transformer for Learning to Fine-tune Video Foundation Models