AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Learning Taxonomic Trees with Hierarchical Representation Regularization for Large Multimodal Models

Hulingxiao He, Zhi Tan, Yuxin Peng

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Layout-aware document understanding · 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 HiR2 regularizer to SFTQwen2.5-VL-3B SFT · CUB-200 base classesHCA: 9.36 -> 35.38Improved
add HiR2 regularizer to SFTQwen2.5-VL-3B SFT · iNat-Animal base classesHCA: 14.06 -> 18.07Improved
add HiR2 regularizer to DFTQwen2.5-VL-3B DFT · iNat-Plant base classesHCA: 8.15 -> 11.66Improved
add HiR2 regularizer to SFT on other LMMsIntern3.5-VL-1B SFT · iNat-Animal base classesHCA: 2.04 -> 3.04Improved
add HiR2 regularizer to SFT on LLaVA-OV-1.5-4BLLaVA-OV-1.5-4B SFT · iNat-Animal novel classesHCA: 6.05 -> 6.09No clear effect
add entailment loss to SFT baseline (ablation)HiR2 · iNat-Animal base classesHCA: 14.06 -> 17.89Improved
add dispersive loss on top of entailment loss (ablation)HiR2 · iNat-Animal base classesHCA: 17.89 -> 18.07Improved
add dispersive loss on top of entailment loss (ablation)HiR2 · iNat-Animal base classesleaf accuracy: 40.78 -> 41.37Improved
use spherical angular dispersive loss instead of hyperbolic geodesic variant (ablation)HiR2 dispersive loss · iNat-Animal base classesHCA: 15.33 -> 18.07Improved
use spherical angular dispersive loss instead of tangent-space variants (ablation)HiR2 dispersive loss · iNat-Animal base classesleaf accuracy: 40.67 -> 41.37 (vs variant III)Improved
regularize last layer (28) instead of layer 21 (ablation)HiR2 · iNat-Animal base classesHCA: 11.32 -> 18.07Improved
regularize all layers instead of only the last layer (ablation)HiR2 · iNat-Animal base classesHCA: 18.07 -> 12.06Hurt

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