AI Paper Atlas

Research map / Continual learning / Domain and prompt-based

ADEPT: Continual Pretraining via Adaptive Expansion and Dynamic Decoupled Tuning

Jinyang Zhang, Yue Fang, Hongxin Ding, Weibin Liao, Muyang Ye, Junfeng Zhao, Yasha Wang, Xu Chu

ICLR 2026 · Continual learning › Domain and prompt-based · approach: Domain-incremental learning with memory

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add General-Competence Guided Selective Layer Expansion (Stage 1) (ablation)ADEPT · Medical CPT, Qwen3-1.7BMMLU accuracy: 57.31 -> 62.80Improved
add Adaptive Unit-Wise Decoupled Tuning (Stage 2) (ablation)ADEPT · Medical CPT, Qwen3-1.7BMedQA accuracy: 49.23 -> 50.75Improved
use importance-guided expansion instead of uniform layer insertion (ablation)ADEPT · Medical CPT, Llama3-8BMMLU accuracy: 64.19 -> 65.17Improved
importance-guided expansion plus decoupled tuning versus full-parameter CPTQwen3-1.7B-Base math CPT · GSM8Kaccuracy: 51.86 -> 70.51Improved
use ADEPT instead of full-parameter CPT to reduce forgettingQwen3-8B-Base math CPT · MMLUaccuracy: 74.90 -> 76.80Improved
add data replay to full-parameter CPTPT-Full · Qwen3-1.7B math CPT, MMLUaccuracy: 60.07 -> 60.69No clear effect
use LoRA CPT instead of full-parameter CPTPT-LoRA · Qwen3-1.7B medical CPT, MMLUaccuracy: 59.44 -> 57.31Hurt
update only a small subset of parameters in expanded layers (ADEPT)training time · Medical CPT, Qwen3-8Btraining time: 8 days, 9h -> 3 days, 15hImproved
ADEPT versus Llama Pro on Llama3-8B medical CPTLlama3-8B-Base · CMBaccuracy: 47.05 -> 61.78Improved
expand layers before decoupling parameters (ablation)Qwen3-8B-Base · domain activation separation (KDE)activation overlapImproved
ADEPT versus full pretraining for token shiftmedical CPT · token distribution shiftshare of shifted tokens: 5.61% -> 2.18%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy