Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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

About

Multimodal In-Context Learning (ICL) has emerged as a practical inference paradigm for Multimodal Large Language Models, where a small set of interleaved image-text In-Context Demonstrations (ICDs) conditions the model to solve new tasks. Despite its flexibility, multimodal ICL incurs high inference latency and suffers from instability due to sensitivity to demonstration formatting, ordering, and content. To address these limitations, we propose Hyper-ICL, a lightweight, training-based framework for demonstration-free multimodal ICL that reconstructs demonstration effects directly without requiring ICDs at inference time. Hyper-ICL learns a parameter-efficient low-rank logit-level adapter that calibrates attention distributions to better match demonstration-induced attention redistribution. To capture how demonstration influence varies across queries, we introduce a query-adaptive modulation mechanism that adaptively controls intervention strength at token level across layers and heads based on the current query. Finally, we propose a layer-wise hyperbolic anchor distillation loss that aligns intermediate student features to a demonstration-conditioned teacher via Lorentz geodesic distance. This loss encourages the student to reconstruct the demonstration-query relationships induced by ICDs. Extensive experiments across six different multimodal benchmarks (including VQAv2, OK-VQA, and COCO Caption) demonstrate that Hyper-ICL consistently improves accuracy and stability over vanilla ICL and existing state-of-the-art methods.

Niloufar Alipour Talemi, Hossein Kashiani, Fatemeh Afghah• 2026

Related benchmarks

TaskDatasetResultRank
Multimodal UnderstandingSEED-Bench
Accuracy48.31
571
Visual Question AnsweringOK-VQA
Accuracy75.32
331
Visual Question AnsweringVQA v2
Accuracy82.24
257
Multimodal UnderstandingMME
Score82.13
150
Visual Question AnsweringVQA 10,000 samples v2 (val)
Accuracy (VQA)71.17
30
Caption Hallucination AnalysisCOCO
CHAIRs8.29
21
Image CaptioningCOCO full (val)
CIDEr135.7
18
Visual Question AnsweringOK-VQA full (val)
Accuracy62.24
18
Image CaptioningFlickr30K
Score93.79
6
Showing 9 of 9 rows

Other info

Follow for update