AI Paper Atlas

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

Hyperbolic Multimodal Continual Learning

Jiahong Liu, Ming Shen, Xiaohao Liu, Rex Ying, Menglin Yang, Tat-Seng Chua, Irwin King

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
Use HMCL instead of Vanilla fine-tuningMERU-L · Continual multimodal classification and retrievalOverall performance: 38.22 -> 41.36Improved
Use HMCL instead of Vanilla fine-tuningMERU-L · Continual multimodal classification and retrievalClassification BWT: -6.46 -> -0.74Improved
Use HMCL instead of Vanilla fine-tuningMERU-B · Continual multimodal classification and retrievalOverall performance: 41.82 -> 44.21Improved
Use HMCL instead of Vanilla fine-tuningHyCoCLIP-B · Continual multimodal classification and retrievalRetrieval image-to-text R@1: 45.34 -> 57.55Improved
Use HMCL instead of Vanilla fine-tuningHyCoCLIP-B · Continual multimodal classification and retrievalOverall BWT: -5.42 -> -0.43Improved
Add EWC to Vanilla fine-tuningMERU-L · Continual multimodal classification and retrievalOverall performance: 38.22 -> 38.22No clear effect
Add GEM to Vanilla fine-tuningMERU-B · Continual multimodal classification and retrievalRetrieval R@5: 33.66 -> 29.06Hurt
Add C-FLAT to Vanilla fine-tuningMERU-L · Continual multimodal classification and retrievalOverall performance: 38.22 -> 37.16Hurt
Add C-FLAT to Vanilla fine-tuningHyCoCLIP-B · Continual multimodal classification and retrievalOverall performance: 42.33 -> 42.32No clear effect
Add C-FLAT to Vanilla fine-tuningMERU-L · Continual multimodal classification and retrievalRetrieval BWT: -6.46 -> -2.56Improved
Use HMCL instead of Vanilla on old-task driftHMCL across datasets · Continual multimodal classification and retrievalRadial, angular, cross-modal, paired-distance drift: 61.0%, 57.8%, 73.8%, 59.5% reductionsImproved
Add block-wise constraint to HMCL updates (ablation)HMCL · Continual multimodal classification and retrievalBackward transfer stabilityImproved

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