Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

SEA: Hierarchically searching efficient adapters for pre-trained models.

Shun Lu¹, Fangyuan Mao², Junkun Chen³

¹Research Center for Intelligent Computing Systems, Institute of Computing Technology, SKLP, Chinese Academy of Sciences, Beijing, 100190, China; Kuaishou Technology, 100000, China; Beijing Institute of Control Engineering, Beijing, 100049, China.

Neural Networks : the Official Journal of the International Neural Network Society

|April 10, 2026

Summary

Related Concept Videos

Improving Translational Accuracy

Improving Translational Accuracy

Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...

Improving Translational Accuracy

Improving Translational Accuracy

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Acupuncture inhibits autophagy and promotes synaptic repair by activating the PI3K/Akt/mTOR pathway in a rat model of depression in Parkinson's disease.

Acupuncture in medicine : journal of the British Medical Acupuncture Society·2026

Same author

Atorvastatin promotes lipid catabolism in colorectal cancer via FDFT1‒mediated inhibition of the PI3K/AKT pathway.

Cell communication and signaling : CCS·2026

Same author

Analysis of genetic mutation distribution and metabolic characteristics in patients with primary carnitine deficiency from the Ganzhou area, China.

Clinica chimica acta; international journal of clinical chemistry·2026

Same author

Angelica sinensis polysaccharides prevents marrow cell apoptosis during stress via regulating endoplasmic reticulum-mitochondrion and adaptive metabolism.

Journal of ethnopharmacology·2025

Same author

Molecular identification and phenotypic study of a novel <i>HBB</i>: c.-23A>G mutation in the 5' untranslated region.

Frontiers in medicine·2025

Same author

Complete mole with co-existing fetus: a rare case report of full-term delivery with favorable maternal and fetal outcomes.

Taiwanese journal of obstetrics & gynecology·2025

Same journal

A practical design of backdoor trigger under frequency-based orthogonality constraints.

Neural networks : the official journal of the International Neural Network Society·2026

Same journal

EEG fine-grained visual semantic decoding via a multimodal framework.

Neural networks : the official journal of the International Neural Network Society·2026

Same journal

Collaborative-adversarial jailbreaking: A propagation-aware attack framework for multi-agent code generation systems.

Neural networks : the official journal of the International Neural Network Society·2026

Same journal

Theoretical analysis of the denoising autoencoder using Tweedie's formula.

Neural networks : the official journal of the International Neural Network Society·2026

Same journal

Frequency-based cross-attention fusion network for RGB-D salient object detection.

Neural networks : the official journal of the International Neural Network Society·2026

Same journal

HTNet: A self-supervised heterogeneous triple network for multi-modal data.

Neural networks : the official journal of the International Neural Network Society·2026

See all related articles

This summary is machine-generated.

Search Efficient Adapters (SEA) optimizes pre-trained models by systematically searching for efficient adapter structures. This approach enhances generalizability and avoids extra inference overhead, outperforming existing methods on various tasks.

Area of Science:

Artificial Intelligence
Machine Learning
Computer Vision

Background:

Adapting large-scale pre-trained models is crucial for downstream task performance.
Hand-crafted adapters can limit generalizability and increase inference overhead.
Existing neural architecture search (NAS) for adapters often overlooks search space design and supernet training bias.

Purpose of the Study:

To propose a general framework, Search Efficient Adapters (SEA), for designing efficient and generalizable adapters.
To address limitations in existing NAS techniques for adapter design.
To enable seamless integration of searched adapters without additional inference overhead.

Main Methods:

SEA introduces a hierarchical adapter search space with refined granularity for depths, widths, and channel combinations.

Keywords:

Adapter Hierarchical search Neural architecture search

Related Experiment Videos

A sample-efficient supernet training mechanism, inspired by Upper Bound Confidence (UCB) theory, is implemented to mitigate training bias.

The framework systematically searches for diverse adapter structures to enhance generalizability.

Main Results:

SEA achieved 0.53% higher average accuracy than GLoRA across 19 few-shot tasks on VTAB-1k.
When adapted on COCO, SEA improved mean average precision by 7.3% (detection) and 5.6% (segmentation) over SSF.
The study revealed insights into the relationship between adapter structures and downstream task performance.

Conclusions:

SEA offers a superior method for designing efficient and generalizable adapters for pre-trained models.
The hierarchical search space and sample-efficient training mechanism contribute to improved performance and reduced overhead.
SEA provides valuable insights for future research in adapter design and optimization.