AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

TRACER: Persistent Regularization for Robust Multimodal Finetuning

Hesam Asadollahzadeh, Feng Liu, Christopher Leckie, Sarah Monazam Erfani

ICML 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Mixture-of-experts for multimodal continual tuning · 0 citations (OpenAlex)

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
replace static SD anchor with WMA teacher trajectory regularization (ablation)Static SD · ImageNet distribution shifts (ViT-B/16)avg shifts accuracy: 59.55 -> 65.46Improved
add static self-distillation to direct finetuningDirect FT · ImageNet distribution shifts (ViT-B/16)avg shifts accuracy: 57.08 -> 59.55Improved
finetune CLIP directly on ImageNetCLIP ViT-B/16 zero-shot · ImageNet distribution shifts (ViT-B/16)avg shifts accuracy: 59.46 -> 57.08Hurt
add CRD loss alone to contrastive finetuning (ablation)TRACER base · ImageNet 5-shift avgOOD accuracy: 59.40 -> 63.12Improved
add FD loss alone to contrastive finetuning (ablation)TRACER base · ImageNet 5-shift avgOOD accuracy: 59.40 -> 63.52Improved
add ICL loss alone to contrastive finetuning (ablation)TRACER base · ImageNet 5-shift avgOOD accuracy: 59.40 -> 59.95No clear effect
add CrossKD and ICL on top of FD+CRD (ablation)TRACER · ImageNet 5-shift avgOOD accuracy: 64.08 -> 64.06No clear effect
raise distillation strength lambda_SD from 0.1 to 1.5 (ablation)TRACER · ImageNet 5-shift avgOOD accuracy: 61.86 -> 64.12Improved
raise distillation strength lambda_SD from 1.5 to 10.0 (ablation)TRACER · ImageNet 5-shift avgOOD accuracy: 64.12 -> 63.27Hurt
raise distillation strength lambda_SD from 0.1 to 5.0 (ablation)TRACER · ImageNet 5-shift avgOOD ECE: 0.1356 -> 0.0670Improved
update WMA teacher every 100 steps instead of every step (ablation)TRACER · ImageNet 5-shift avgOOD accuracy: 64.07 -> 64.22No clear effect
use dynamic WMA distillation instead of direct finetuningDirect FT · Colored MNIST toyforgetting rate on MNIST: 37.9% -> 0.1%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