We are delighted to announce that 2 papers from IIPL (Prof. YoungBin Kim) have been accepted to CIKM 2026.

Title: 

Beyond One-Shot Expansion: Contrastive Evidence Exploration for Multi-Hop Retrieval

Authors: 

JungMin Yun, YoungBin Kim

Abstract: 

Multi-hop question answering requires retrieving evidence linked through intermediate facts, but one-shot query expansion can miss needed information or introduce redundant passages. This training-free retrieval framework iteratively examines retrieved evidence and generates new probes for unresolved information needs. It also builds passage-specific contrastive facets during indexing to distinguish near-neighbor documents and uses coverage-aware selection to assemble complementary evidence. Evaluations on MuSiQue, HotpotQA, and 2WikiMultihopQA report gains in retrieval and downstream question answering.

Title: 

RI-Prune: Relation and Information Aware Visual Token Pruning for Multi-Image Vision-Language Models

Authors: 

Dongjin Lee, Eunju Lee, Jinhee Jang, Yoonji Lee, YoungBin Kim

Abstract: 

Multi-image vision-language tasks produce large visual token sequences, and attention-based pruning methods developed for single images can discard important information. RI-Prune uses training-free, attention-free token selection before the language decoder. It combines relevance to the text query, correspondence among images, and the representativeness of tokens within each image, with balanced selection across images. The researchers also use token ablation tests to expose weaknesses in conventional attention scores. Across several models and multi-image benchmarks, the method improves the trade-off between token efficiency and task performance.