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

While directly fine-tuning (FT) large-scale, pretrained models ontask-specific data is well-known to induce strong in-distribution taskperformance, recent works have demonstrated that different adaptationprotocols, such as linear probing (LP) prior to FT, can improveout-of-distribution generalization. However, the design space of suchadaptation protocols remains under-explored and the evaluation of suchprotocols has primarily focused on distribution shifts. Therefore, in thiswork, we evaluate common adaptation protocols across distributions shifts andmachine learning safety metrics (e.g., anomaly detection, calibration,robustness to corruptions). We find that protocols induce disparate trade-offsthat were not apparent from prior evaluation. Further, we demonstrate thatappropriate pairing of data augmentation and protocol can substantiallymitigate this trade-off. Finally, we hypothesize and empirically see that usinghardness-promoting augmentations during LP and then FT with augmentations maybe particularly effective for trade-off mitigation.

Exploring the Design of Adaptation Protocols for Improved Generalization and Machine Learning Safety