AI Paper Atlas

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

Octopus: History-Free Gradient Orthogonalization for Continual Learning in Multimodal Large Language Models

Yuehao Liu, Shanyan Guan, Weijia Zhang, Xuanming Shang, Yanhao Ge, Wei Li, Chao Ma

CVPR 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
add two-stage finetuning (ablation)Octopus · UCITLast: 61.18 -> 71.01Improved
add two-stage finetuning (ablation)Octopus · UCITImd. average: 62.47 -> 70.60Improved
add norm regularization (lambda 1e-2) (ablation)Octopus · UCITLast: 64.77 -> 71.01Improved
add norm regularization (lambda 1e-2) (ablation)Octopus · UCITImd. average: 73.36 -> 70.60Hurt
add norm regularization (lambda 1e-2) (ablation)Octopus · UCITBWT: -8.59 -> +0.41Improved
use GPWC orthogonality instead of previous-parameter orthogonality (ablation)Octopus · UCITLast: 66.71 -> 71.01Improved
use GPWC orthogonality instead of previous-parameter orthogonality (ablation)Octopus · UCITBWT: -2.51 -> +0.41Improved
use GPWC instead of gradients on previous task data (ablation)Octopus · UCITLast: 62.25 -> 71.01Improved
use GPWC instead of gradients on previous task data (ablation)Octopus · UCITBWT: -8.95 -> +0.41Improved
add parameter orthogonality on top of GPWC (ablation)Octopus · UCITLast: 71.01 -> 71.04No clear effect
change task order from R-A-V-I-C-F (ablation)Octopus · UCITLast: 71.01 -> 69.04No clear effect
use Octopus instead of O-LoRALLaVA-v1.5-7B · CoINLast: 60.77 -> 65.10Improved

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