@inproceedings{chang-li-2026-learning,
title = "Learning to Evict: Policy-Driven {KV} Cache Optimization for Multi-Hop Multimodal Spatial Reasoning",
author = "Chang, Jiale and
Li, Ying",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.729/",
pages = "16932--16947",
ISBN = "979-8-89176-459-0",
abstract = "Visualization-of-thought (VoT) has substantially advanced multimodal spatial reasoning by interleaving visual thinking with verbal reasoning. However, multi-hop VoT reasoning introduces a fundamental scalability bottleneck: the autoregressive accumulation of interleaved visual and textual tokens causes KV cache memory to grow rapidly across successive hops. We present ScoutKV, a framework that recasts KV cache eviction as a reinforcement learning problem. ScoutKV trains a dedicated lightweight policy agent per KV head, conditioned on key{--}value representations, cross-hop positional encodings, and modality embeddings, to predict each token{'}s long-term utility across the full reasoning trajectory. Training combines offline supervised fine-tuning with future-query attention targets and online GRPO optimization with a multimodal process reward targeting low-entropy text anchors and modality-specific spatial anchors. ScoutKV consistently reduces FLOPs, latency, and peak memory while maintaining or improving accuracy across multimodal spatial reasoning benchmarks, and generalizes robustly to broader vision-language tasks."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chang-li-2026-learning">
<titleInfo>
<title>Learning to Evict: Policy-Driven KV Cache Optimization for Multi-Hop Multimodal Spatial Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiale</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ying</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Visualization-of-thought (VoT) has substantially advanced multimodal spatial reasoning by interleaving visual thinking with verbal reasoning. However, multi-hop VoT reasoning introduces a fundamental scalability bottleneck: the autoregressive accumulation of interleaved visual and textual tokens causes KV cache memory to grow rapidly across successive hops. We present ScoutKV, a framework that recasts KV cache eviction as a reinforcement learning problem. ScoutKV trains a dedicated lightweight policy agent per KV head, conditioned on key–value representations, cross-hop positional encodings, and modality embeddings, to predict each token’s long-term utility across the full reasoning trajectory. Training combines offline supervised fine-tuning with future-query attention targets and online GRPO optimization with a multimodal process reward targeting low-entropy text anchors and modality-specific spatial anchors. ScoutKV consistently reduces FLOPs, latency, and peak memory while maintaining or improving accuracy across multimodal spatial reasoning benchmarks, and generalizes robustly to broader vision-language tasks.</abstract>
<identifier type="citekey">chang-li-2026-learning</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.729/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>16932</start>
<end>16947</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Learning to Evict: Policy-Driven KV Cache Optimization for Multi-Hop Multimodal Spatial Reasoning
%A Chang, Jiale
%A Li, Ying
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F chang-li-2026-learning
%X Visualization-of-thought (VoT) has substantially advanced multimodal spatial reasoning by interleaving visual thinking with verbal reasoning. However, multi-hop VoT reasoning introduces a fundamental scalability bottleneck: the autoregressive accumulation of interleaved visual and textual tokens causes KV cache memory to grow rapidly across successive hops. We present ScoutKV, a framework that recasts KV cache eviction as a reinforcement learning problem. ScoutKV trains a dedicated lightweight policy agent per KV head, conditioned on key–value representations, cross-hop positional encodings, and modality embeddings, to predict each token’s long-term utility across the full reasoning trajectory. Training combines offline supervised fine-tuning with future-query attention targets and online GRPO optimization with a multimodal process reward targeting low-entropy text anchors and modality-specific spatial anchors. ScoutKV consistently reduces FLOPs, latency, and peak memory while maintaining or improving accuracy across multimodal spatial reasoning benchmarks, and generalizes robustly to broader vision-language tasks.
%U https://aclanthology.org/2026.emnlp-main.729/
%P 16932-16947
Markdown (Informal)
[Learning to Evict: Policy-Driven KV Cache Optimization for Multi-Hop Multimodal Spatial Reasoning](https://aclanthology.org/2026.emnlp-main.729/) (Chang & Li, EMNLP 2026)
ACL