论文检索

Computer Vision · 3D Rendering & Reconstruction

Yuanhang Lei, Boming Zhao, Zesong Yang, Xingxuan Li, Tao Cheng, Haocheng Peng, Ru Zhang, yang yang, Siyuan Huang, Yujun Shen 等

Modeling wind-driven object dynamics from video observations is highly challenging due to the invisibility and spatio–temporal variability of wind, as well as the complex deformations of objects. We present DiffWind, a physics-informed differentiable framework that unifies wind–object interaction modeling, video-based reconstruction, and forward simulation. Specifically, we represent wind as a grid-based physical field and objects as particle systems derived from 3D Gaussian Splatting, with their interaction modeled by the Material Point Method (MPM). To recover wind-driven object dynamics, we introduce a reconstruction framework that jointly optimizes the spatio–temporal wind force field and object motion through differentiable rendering and simulation. To ensure physical validity, we incorporate the Lattice Boltzmann Method (LBM) as a physics-informed constraint, enforcing compliance with fluid dynamics laws. Beyond reconstruction, our method naturally supports forward simulation under novel wind conditions and enable new applications such as wind retargeting. We further introduce WD-Objects, a dataset of synthetic and real-world wind-driven scenes. Extensive experiments demonstrate that our method significantly outperforms prior dynamic scene modeling approaches in both reconstruction accuracy and simulation fidelity, opening a new avenue for video-based wind–object interaction modeling. The project page is available at: [https://zju3dv.github.io/DiffWind/](https://zju3dv.github.io/DiffWind/).

Deep Learning · Generative Models and Autoencoders

Bedionita Soro, Aoxuan Zhang, Bruno Andreis, Jaehyeong Jo, Song Chong, Sung Ju Hwang

Model merging in weight space is an efficient way to reuse pretrained models, but existing methods typically assume matching architectures or sizes, making heterogeneous merges brittle or infeasible. We address this limitation by encoding model weights into a smooth latent space, enabling cross-architecture operations, and performing the merge in the latent space before decoding back to weights. This approach faces two major challenges. First, LLMs contain billions of parameters, which makes latent encoding computationally demanding. Second, using high compression ratios often hinders the encoder’s ability to generalize to unseen weights. We tackle these issues with a transformer-based variational autoencoder (VAE) trained in a two-stage compression curriculum with structured layer-aware chunking: the model first learns a high-capacity latent representation and then distills to a compact code, improving both stability and out-of-distribution generalization. To align heterogeneous models, we introduce a dimensionality-matching projection that allows interpolation between models of different sizes. Empirically, latent-space interpolation is consistently more robust than direct weight-space averaging and yields stronger downstream performance when merging models of different sizes. Together, these components provide a scalable, architecture-agnostic recipe for model merging.

Computer Vision · Image and Video Generation

Hengyuan Xu, Wei Cheng, Peng Xing, Yixiao Fang, Shuhan Wu, Rui Wang, Xianfang Zeng, Daxin Jiang, Gang Yu, Xingjun Ma 等

Identity-consistent (ID-consistent) generation has become an important focus in text-to-image research, with recent models achieving notable success in producing images aligned with a reference identity. Yet, the scarcity of large-scale paired datasets—containing multiple images of the same individual—forces most approaches to adopt reconstruction-based training. This reliance often leads to a failure mode we term copy-paste, where the model directly replicates the reference face rather than preserving identity across natural variations in pose, expression, or lighting. Such over-similarity undermines controllability and limits the expressive power of generation. To address these limitations, we (1) construct a large-scale paired dataset, MultiID-2M, tailored for multi-person scenarios, providing diverse references for each identity; (2) introduce a benchmark that quantifies both copy-paste artifacts and the trade-off between identity fidelity and variation; and (3) propose a novel training paradigm with a contrastive identity loss that leverages paired data to balance fidelity with diversity. These contributions culminate in WithAnyone, a diffusion-based model that effectively mitigates copy-paste while preserving high identity similarity. Extensive experiments—both qualitative and quantitative—demonstrate that WithAnyone substantially reduces copy-paste artifacts, improves controllability over pose and expression, and maintains strong perceptual quality. User studies further validate that our method achieves high identity fidelity while enabling expressive, controllable generation.

Reinforcement Learning · Deep RL

Yunyue Wei, Chenhui Zuo, Yanan Sui

Controlling high-dimensional biological and robotic systems is challenging due to expansive state–action spaces, where effective exploration is critical. Commonly used exploration strategies in reinforcement learning are largely undirected with sharp degradation as action dimensionality grows. Many existing methods resort to dimensionality reduction, which constrains policy expressiveness and forfeits system flexibility. We introduce Q-guided Flow Exploration (Qflex), a scalable reinforcement learning method that conducts exploration directly in the native high-dimensional action space. During training, Qflex traverses actions from a learnable source distribution along a probability flow induced by the learned value function, aligning exploration with task-relevant gradients rather than isotropic noise. Our proposed method substantially outperforms representative online reinforcement learning baselines across diverse high-dimensional continuous-control benchmarks. Qflex also successfully controls a whole-body human musculoskeletal model to perform agile, complex movements, demonstrating superior scalability and sample efficiency in very high-dimensional settings. Our results indicate that value-guided flows offer a principled and practical route to exploration at scale.

Applications · Health

Manuela González-González, Soufiane Belharbi, Muhammad Zeeshan, Masoumeh Sharafi, Muhammad Haseeb Aslam, Marco Pedersoli, Alessandro Lameiras Koerich, Simon L Bacon, Eric Granger

Ambivalence and hesitancy (A/H), closely related constructs, are the primary reasons why individuals delay, avoid, or abandon health behaviour changes. They are subtle and conflicting emotions that sets a person in a state between positive and negative orientations, or between acceptance and refusal to do something. They manifest as a discord in affect between multiple modalities or within a modality, such as facial and vocal expressions, and body language. Although experts can be trained to recognize A/H as done for in-person interactions, integrating them into digital health interventions is costly and less effective. Automatic A/H recognition is therefore critical for the personalization and cost-effectiveness of digital behaviour change interventions. However, no datasets currently exist for the design of machine learning models to recognize A/H. This paper introduces the Behavioural Ambivalence/Hesitancy (BAH) dataset collected for multimodal recognition of A/H in videos. It contains 1,427 videos with a total duration of 10.60 hours, captured from 300 participants across Canada, answering predefined questions to elicit A/H. It is intended to mirror real-world digital behaviour change interventions delivered online. BAH is annotated by three experts to provide timestamps that indicate where A/H occurs, and frame- and video-level annotations with A/H cues. Video transcripts, cropped and aligned faces, and participant metadata are also provided. Since A and H manifest similarly in practice, we provide a binary annotation indicating the presence or absence of A/H. Additionally, this paper includes benchmarking results using baseline models on BAH for frame- and video-level recognition, zero-shot prediction, and personalization with source-free domain adaptation methods. The limited performance highlights the need for adapted multimodal and spatio-temporal models for A/H recognition. Results obtained with specialized fusion methods are shown to assess the presence of conflicts between modalities, additionally temporal modelling for within-modality conflicts are essential for more discriminant A/H recognition. The data, code, and pretrained weights are publicly available: https://github.com/LIVIAETS/bah-dataset.

Computer Vision · Classification and Understanding

Masoumeh Sharafi, Soufiane Belharbi, Muhammad Zeeshan, HOUSSEM Salem, Ali Etemad, Alessandro Lameiras Koerich, Marco Pedersoli, Simon L Bacon, Eric Granger

Facial expression recognition (FER) models are employed in many video-based affective computing applications, such as human-computer interaction and healthcare monitoring. However, deep FER models often struggle with subtle expressions and high inter-subject variability, limiting their performance in real-world applications. To improve performance, source-free domain adaptation (SFDA) methods have been proposed to personalize a pretrained source model using only unlabeled target domain data, thereby avoiding data privacy, storage, and trans- mission constraints. This paper addresses a common challenging scenario where source data is unavailable for adaptation, and only unlabeled target data consisting solely of neutral expressions is available. SFDA methods are not typically designed to adapt using target data from only a single class. Further, using models to generate facial images with non-neutral expressions can be unstable and computationally intensive. In this paper, the Source-Free Domain Adaptation with Personalized Feature Translation (SFDA-PFT) method is proposed for SFDA. Unlike current image translation methods for SFDA, our lightweight method op- erates in the latent space. We first pre-train the translator on source domain data to transform the subject-specific style features from one source subject into another. Expression information is preserved by optimizing a combination of expression consistency and style-aware objectives. Then, the translator is adapted to neutral target data, without using source data or image synthesis. By translating in the latent space, SFDA-PFT avoids the complexity and noise of face expression generation, producing discriminative embeddings optimized for classification. Using SFDA-PFT eliminates the need for image synthesis, reduces computational overhead, and only adapts a lightweight translator, making the method efficient compared to image-based translation. Our extensive experiments on four challenging video FER benchmark datasets, BioVid, stressID, BAH, and Af-Wild2, show that PFT consistently outperforms state-of-the-art SFDA methods, providing a cost-effective approach that is suitable for real-world, privacy-sensitive FER applications. Our code is publicly available at: github.com/MasoumehSharafi/SFDA-PFT.

Deep Learning · Graph Neural Networks

Yancheng Wang, Chengshuai Zhao, Dongfang Sun, huan liu, Yingzhen Yang

In this paper, we propose a novel node-level graph diffusion method with low-rank feature learning for few-shot node classification (FSNC), termed Low-Rank Few-Shot Graph Diffusion Model or LR-FGDM. LR-FGDM first employs a novel Few-Shot Graph Diffusion Model (FGDM) as a node-level graph generative method to generate an augmented graph with an enlarged support set, then performs low-rank transductive classification to obtain the few-shot node classification results. Our graph diffusion model, FGDM, comprises two components, the Hierarchical Graph Autoencoder (HGAE) with an efficient hierarchical edge reconstruction method and a new prototypical regularization, and the Latent Diffusion Model (LDM). The low-rank regularization is robust to the noise inherently introduced by the diffusion model and empirically inspired by the Low Frequency Property. We also provide a strong theoretical guarantee justifying the low-rank regularization for the transductive classification in few-shot learning. To further enhance the performance of LR-FGDM, we introduce LRA-LR-FGDM with a novel efficient LR-Attention layer, or the LRA layer, which applies self-attention to the output of the LR-FGDM encoder. The LRA layer further reduces the kernel complexity of LR-FGDM and contributes to a tighter generalization bound, leading to improved performance. Extensive experimental results evidence the effectiveness of LR-FGDM for few-shot node classification, which outperforms the current state-of-the-art. The code of the LR-FGDM is available at \url{https://github.com/Statistical-Deep-Learning/LR-FGDM}.

Computer Vision · Image and Video Generation

Zhikai Li, jiatong li, Xuewen Liu, Wangbo Zhao, Pan Du, Kaicheng Zhou, Qingyi Gu, Yang You, Zhen Dong, Kurt Keutzer

The rapid development of visual generative models raises the need for more scalable and human-aligned evaluation methods. While the crowdsourced Arena platforms offer human preference assessments by collecting human votes, they are costly and time-consuming, inherently limiting their scalability. Leveraging vision-language model (VLMs) as substitutes for manual judgments presents a promising solution. However, the inherent hallucinations and biases of VLMs hinder alignment with human preferences, thus compromising evaluation reliability. Additionally, the static evaluation approach lead to low efficiency. In this paper, we propose K-Sort Eval, a reliable and efficient VLM-based evaluation framework that integrates posterior correction and dynamic matching. Specifically, we curate a high-quality dataset from thousands of human votes in K-Sort Arena, with each instance containing the outputs and rankings of K models. When evaluating a new model, it undergoes (K+1)-wise free-for-all comparisons with existing models, and the VLM provide the rankings. To enhance alignment and reliability, we propose a posterior correction method, which adaptively corrects the posterior probability in Bayesian updating based on the consistency between the VLM prediction and human supervision. Moreover, we propose a dynamic matching strategy, which balances uncertainty and diversity to maximize the expected benefit of each comparison, thus ensuring more efficient evaluation. Extensive experiments show that K-Sort Eval delivers evaluation results consistent with K-Sort Arena, typically requiring fewer than 90 model runs, demonstrating both its efficiency and reliability. The dataset and code are publicly available.

Computer Vision · Image and Video Generation

Hongyu Zhang, Yufan Deng, Zilin Pan, Peng-Tao Jiang, Bo Li, Qibin Hou, Zhen Dong, Zhiyang Dou, Zhou Daquan

Generating high-quality videos from complex temporal descriptions, which refer to prompts containing multiple sequential actions, remains a significant challenge. Existing methods are constrained by an inherent trade-off: using multiple short prompts fed sequentially into the model improves action fidelity but compromises temporal consistency, while a single complex prompt preserves consistency at the cost of prompt following capability. We attribute this problem to two primary causes: temporal misalignment between video content and the prompt, and conflicting attention coupling between motion-related visual objects and their associated text conditions. To address these challenges, we propose a novel, training-free attention mechanism, Temporal-wise Separable Attention (TS-Attn), which dynamically rearranges attention distribution to ensure temporal awareness and global coherence in multi-event scenarios. TS-Attn can be seamlessly integrated into various pre-trained text-to-video models, boosting StoryEval-Bench scores by 33.5% and 16.4% on Wan2.1-T2V-14B and Wan2.2-T2V-A14B with only a 2% increase in inference time. It also supports plug-and-play usage across models for multi-event image-to-video generation. The source code and video demos are available in the supplementary materials.

Deep Learning · Generative Models and Autoencoders

Qinyuan Wu, Soumi Das, Mahsa Amani, Bishwamittra Ghosh, Mohammad Aflah Khan, Krishna Gummadi, Muhammad Bilal Zafar

Rote learning is a memorization technique based on repetition. Many researchers argue that rote learning hinders generalization because it encourages verbatim memorization rather than deeper understanding. This concern extends even to factual knowledge, which inevitably requires a certain degree of memorization. In this work, we challenge this view and demonstrate that large language models (LLMs) can, in fact, generalize over rote memorized data. We introduce a two-phase “memorize-then-generalize” framework, where the model first rote memorizes factual subject-object associations using a synthetic semantically meaningless key token and then learns to generalize by fine-tuning on a small set of semantically meaningful prompts. Extensive experiments over 8 LLMs show that the models can reinterpret rote memorized data through the semantically meaningful prompts, as evidenced by the emergence of structured, semantically aligned latent representations between the key token and the semantically meaningful prompts. This surprising finding opens the door to both effective and efficient knowledge injection as well as possible risks of repurposing the memorized data for malicious usage.

Applications · Everything Else

Zitong Lan, Yiduo Hao, Mingmin Zhao

Audio editing plays a crucial role in VR/AR immersion, virtual conferencing, sound design, and interactive media. However, recent generative audio editing models depend on template-like instruction formats and are restricted to mono-channel audio. Moreover, existing systems require users to specify low-level editing actions, rather than expressing the desired outcome at a higher semantic level. We introduce SmartDJ, a novel framework for stereo audio editing that enables declarative audio editing, where the users describe the desired outcome while delegating the underlying editing operations to the system. Given a high-level instruction, SmartDJ decomposes it into a sequence of atomic edit operations, such as adding, removing, or spatially relocating sound events. These operations are then executed by a diffusion model trained to edit stereo audio. To enable this capability, we design a scalable data synthesis pipeline that produces paired examples of declarative instructions, atomic edit operations, and audios before and after each edit operation. Experiments demonstrate that SmartDJ achieves superior perceptual quality, spatial realism, and semantic alignment compared to prior audio editing methods.

Social Aspects · Accountability, Transparency and Interpretability

Shawn Im, Changdae Oh, Zhen Fang, Sharon Li

Semantic associations such as the link between "bird" and "flew" are foundational for language modeling as they enable models to go beyond memorization and instead generalize and generate coherent text. Understanding how these associations are learned and represented in language models is essential for connecting deep learning with linguistic theory and developing a mechanistic foundation for large language models. In this work, we analyze how these associations emerge from natural language data in attention-based language models through the lens of training dynamics. By leveraging a leading-term approximation of the gradients, we develop closed-form expressions for the weights at early stages of training that explain how semantic associations first take shape. Through our analysis, we reveal that each set of weights of the transformer has closed-form expressions as simple compositions of three basis functions--bigram, token-interchangeability, and context mappings--reflecting the statistics in the text corpus and uncover how each component of the transformer captures the semantic association based on these compositions. Experiments on real-world LLMs demonstrate that our theoretical weight characterizations closely match the learned weights, and qualitative analyses further guide us on how our theorem shines light on interpreting the learned association in transformers.

Optimization · Everything Else

Chenyu Zhou, Tianyi Xu, Jianghao Lin, Dongdong Ge

Large Language Models (LLMs) have shown promising capabilities for solving Operations Research (OR) problems. While reinforcement learning serves as a powerful paradigm for LLM training on OR problems, existing works generally face two key limitations. First, outcome reward suffers from the $\textit{credit assignment problem}$, where correct final answers can reinforce flawed reasoning. Second, conventional discriminative process supervision is $\textit{myopic}$, failing to evaluate the interdependent steps of OR modeling holistically. To this end, we introduce $\textbf{\texttt{StepORLM}}$, a novel self-evolving framework with generative process supervision. At its core, $\texttt{StepORLM}$ features a co-evolutionary loop where a policy model and a generative process reward model (GenPRM) iteratively improve on each other. This loop is driven by a dual-feedback mechanism: definitive, outcome-based verification from an external solver, and nuanced, holistic process evaluation from the GenPRM. The combined signal is used to align the policy via Weighted Direct Preference Optimization (W-DPO) and simultaneously refine the GenPRM. Our resulting 8B-parameter $\texttt{StepORLM}$ establishes a new state-of-the-art across six benchmarks, significantly outperforming vastly larger generalist models, agentic methods, and specialized baselines. Moreover, the co-evolved GenPRM is able to act as a powerful and universally applicable process verifier, substantially boosting the inference scaling performance of both our own model and other existing LLMs. We release our models and code to facilitate future research (https://github.com/0xzhouchenyu/StepORLM).

Social Aspects · Trustworthy Machine Learning

Wendi Li, Changdae Oh, Sharon Li

Optimistic exploration is central to improving sample efficiency in reinforcement learning with human feedback, yet existing exploratory bonus methods often fail to realize true optimism. We provide a theoretical analysis showing that current formulations, under KL or $\alpha$-divergence regularization, unintentionally bias exploration toward high-probability regions of the reference model, thereby reinforcing conservative behavior instead of promoting discovery of uncertain regions. To address this pitfall, we introduce the General Exploratory Bonus (GEB), a novel theoretical framework that provably satisfies the optimism principle. GEB counteracts divergence-induced bias via reference-dependent reward regulation and unifies prior heuristic bonuses as special cases, while extending naturally across the full $\alpha$-divergence family. Empirically, GEB consistently outperforms baselines on alignment tasks across multiple divergence settings and large language model backbones. These results demonstrate that GEB offers both a principled and practical solution for optimistic exploration in RLHF.

Social Aspects · Fairness, Equity, Justice and Safety

Chris Hays, Benjamin Laufer, Solon Barocas, Manish Raghavan

U.S. discrimination law can impose liability on firms that fail to adopt a less discriminatory alternative (LDA), defined as a decision policy that achieves the same business objectives while reducing disparate impact on legally protected groups. Recent scholarship argues that this doctrine has direct implications for algorithmic decision-making in high-stakes domains such as employment, lending, and housing, potentially obligating firms to search for “less discriminatory algorithms” (Black et al., 2024). Regulators have at times encouraged proactive LDA searches, reinforcing the expectation of a good-faith effort to identify equally performant models with lower disparate impact. Model multiplicity makes such searches plausible: retraining with different random seeds can yield models with comparable predictive performance but materially different disparate impacts. Yet firms cannot retrain indefinitely, raising a central question: when is the search sufficient to demonstrate good faith? We formalize LDA search under multiplicity as an optimal stopping problem in which a developer seeks to produce evidence that further search is unlikely to yield meaningful improvements. Our main contribution is an adaptive stopping algorithm that provides a high-probability upper bound on the best disparate-impact gains attainable through continued retraining, enabling developers to certify (e.g., to a court) that additional search is unlikely to help. We also show how stronger distributional assumptions over the model space can yield tighter bounds, and we validate the approach on real-world credit and housing datasets.

Social Aspects · Trustworthy Machine Learning

Lin Long, Changdae Oh, Seongheon Park, Sharon Li

Large vision-language models (LVLMs) achieve strong performance on multimodal tasks, yet they often default to their language prior (LP)---memorized textual patterns from pre-training while under-utilizing visual evidence. Prior analyses of LP mostly rely on input–output probing, which fails to reveal the internal mechanisms governing when and how vision influences model behavior. To address this gap, we present the first systematic analysis of language prior through the lens of chain-of-embedding, which examines the layer-wise representation dynamics within LVLMs. Our analysis reveals a universal phenomenon: each model exhibits a Visual Integration Point (VIP), a critical layer at which visual information begins to meaningfully reshape hidden representations and influence decoding for multimodal reasoning. Building on this observation, we introduce the Total Visual Integration (TVI) estimator, which aggregates representational discrepancy beyond the VIP to quantify how strongly visual query influences response generation. Across 60 model–dataset combinations spanning 10 contemporary LVLMs and 6 benchmarks, we demonstrate that VIP consistently emerges, and that TVI reliably predicts the strength of language prior. This offers a principled toolkit for diagnosing and understanding language prior in LVLMs.

Reinforcement Learning · Multi-agent

Zain Ulabedeen Farhat, Debamita Ghosh, George Atia, Yue Wang

Well-trained multi-agent systems can fail when deployed in real-world environments due to model mismatches between the training and deployment environments, caused by environment uncertainties including noise or adversarial attacks. Distributionally Robust Markov Games (DRMGs) enhance system resilience by optimizing for worst-case performance over a defined set of environmental uncertainties. However, current methods are limited by their dependence on simulators or large offline datasets, which are often unavailable. This paper pioneers the study of online learning in DRMGs, where agents learn directly from environmental interactions without prior data. We introduce the Multiplayer Optimistic Robust Nash Value Iteration (MORNAVI) algorithm and provide the first provable guarantees for this setting. Our theoretical analysis demonstrates that the algorithm achieves low regret and efficiently finds the optimal robust policy for uncertainty sets measured by Total Variation divergence and Kullback-Leibler divergence. These results establish a new, practical path toward developing truly robust multi-agent systems.

Optimization · Learning for Optimization

Ya Shen, Gang Chen, Hui Ma, Mengjie Zhang

Workflow scheduling in cloud computing demands the intelligent allocation of dynamically arriving, graph-structured workflows with varying deadlines onto ever-changing virtual machine resources. However, existing deep reinforcement learning (DRL) schedulers remain limited by rigid, single-path inference architectures that struggle to handle diverse scheduling scenarios. We introduce $\textbf{DEFT}$ ($\textbf{D}$eadline-p$\textbf{E}$rceptive Mixture-o$\textbf{F}$-Exper$\textbf{t}$s), an innovative DRL policy architecture that leverages a specialized mixture of experts, each trained to manage different levels of deadline tightness. To our knowledge, DEFT is the first to introduce and validate a Mixture-of-Experts architecture for dynamic cloud workflow scheduling. By adaptively routing decisions through the most appropriate experts, DEFT is capable of meeting a broad spectrum of deadline requirements that no single expert can achieve. Central to DEFT is a $\textbf{graph-adaptive}$ gating mechanism that encodes workflow DAGs, task states, and VM conditions, using cross-attention to guide expert activation in a fine-grained, deadline-sensitive manner. Experiments on dynamic cloud workflow benchmarks demonstrate that DEFT significantly reduces execution cost and deadline violations, outperforming multiple state-of-the-art DRL baselines.

General Machine Learning · Probabilistic Methods

Paul Hofman, Timo Löhr, Maximilian Muschalik, Yusuf Sale, Eyke Hüllermeier

A reliable representation of uncertainty is essential for the application of modern machine learning methods in safety-critical settings. In this regard, the use of credal sets (i.e., convex sets of probability distributions) has recently been proposed as a suitable approach to representing epistemic uncertainty. However, as with other approaches to epistemic uncertainty, training credal predictors is computationally complex and usually involves (re-)training an ensemble of models. The resulting computational complexity prevents their adoption for complex models such as foundation models and multi-modal systems. To address this problem, we propose an efficient method for credal prediction that is grounded in the notion of relative likelihood and inspired by techniques for the calibration of probabilistic classifiers. For each class label, our method predicts a range of plausible probabilities in the form of an interval. To produce the lower and upper bounds of these intervals, we propose a technique that we refer to as decalibration. Extensive experiments show that our method yields credal sets with strong performance across diverse tasks, including coverage–efficiency evaluation, out-of-distribution detection, and in-context learning. Notably, we demonstrate credal prediction on models such as TabPFN and CLIP—architectures for which the construction of credal sets was previously infeasible.

Deep Learning · Attention Mechanisms

Jinwoo Ahn, Ingyu Seong, Akhil Kedia, Junhan Kim, Hyemi Jang, Kangwook Lee, Yongkweon Jeon

Transformer-based large language models (LLMs) rely on key–value (KV) caching to avoid redundant computation during autoregressive inference. While this mechanism greatly improves efficiency, the cache size grows linearly with the input sequence length, quickly becoming a bottleneck for long‑context tasks. Existing solutions mitigate this problem by evicting prompt KV that are deemed unimportant, guided by estimated importance scores. Notably, a recent line of work proposes to improve eviction quality by “glimpsing into the future”, in which a draft generator produces a surrogate future response approximating the target model's true response, and this surrogate is subsequently used to estimate the importance of cached KV more accurately. However, these approaches rely on computationally expensive draft generation, which introduces substantial prefilling overhead and limits their practicality in real-world deployment. To address this challenge, we propose LookaheadKV, a lightweight eviction framework that leverages the strength of surrogate future response without requiring explicit draft generation. LookaheadKV augments transformer layers with parameter‑efficient modules trained to predict true importance scores with high accuracy. Our design ensures negligible runtime overhead comparable to existing inexpensive heuristics, while achieving accuracy superior to more costly approximation methods. Extensive experiments on long-context understanding benchmarks, across a wide range of models, demonstrate that our method not only outperforms recent competitive baselines in various long-context understanding tasks, but also reduces the eviction cost by up to $14.5$×, leading to significantly faster time-to-first-token. Our code is available at https://github.com/SamsungLabs/LookaheadKV.