AI Paper Atlas

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

Hyper-ICL: Attention Calibration with Hyperbolic Anchor Distillation for Multimodal In-Context Learning

Niloufar Alipour Talemi, Hossein Kashiani, Fatemeh Afghah

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Multimodal retrieval-augmented modeling · 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 Euclidean anchor with hyperbolic anchor distillation (ablation)Hyper-ICL (Idefics-9B) · VQAv2accuracy: 59.64 -> 62.08Improved
replace Euclidean anchor with hyperbolic anchor distillation (ablation)Hyper-ICL (Idefics-9B) · OK-VQAaccuracy: 53.76 -> 55.31Improved
add anchor distillation to supervised loss (ablation)Hyper-ICL (Idefics-9B) · VQAv2accuracy: 57.11 -> 62.08Improved
add supervised loss to hyperbolic anchor loss (ablation)Hyper-ICL (Idefics-9B) · OK-VQAaccuracy: 53.78 -> 55.31Improved
apply module at attention logits instead of post-attention output (ablation)Hyper-ICL (Idefics-9B) · VQAv2accuracy: 58.42 -> 62.08Improved
apply module at attention logits instead of post-attention output (ablation)Hyper-ICL (Idefics-9B) · OK-VQAaccuracy: 50.18 -> 55.31Improved
increase demonstrations from 32 to 64 shots (ablation)Hyper-ICL (Idefics-9B) · OK-VQAaccuracy: 55.31 -> 55.26No clear effect
increase demonstrations from 4 to 32 shots (ablation)Hyper-ICL (Idefics-9B) · VQAv2accuracy: 57.87 -> 62.08Improved
increase adapter rank beyond 4 (ablation)Hyper-ICL (Idefics-9B) · VQAv2accuracy: 62.1 -> 62.2No clear effect
set curvature to 0.1 rather than 0.01 or 1.0 (ablation)hyperbolic anchor distillation · VQAv2 (Idefics-9B)accuracyImproved
replace 32-shot ICL with Hyper-ICL at inferenceIdefics-9B · inference FLOPs and runtimeFLOPs (T) / runtime (ms): 23.364 -> 0.955; 468.55 -> 59.32Improved
replace standard ICL with Hyper-ICLIdefics-9B · COCO caption hallucinationCHAIRs: 16.78 -> 8.29Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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