Yuhan Liu

Yuhan Liu, Hanchen Li, Yihua Cheng et al. · stanford

As large language models (LLMs) take on complex tasks, their inputs are supplemented with longer contexts that incorporate domain knowledge. Yet using long contexts is challenging, as nothing can be generated until the whole context is processed by the LLM. While the context-processing delay can be reduced by reusing the KV cache of a context across different inputs, fetching the KV cache, which contains large tensors, over the network can cause high extra network delays. CacheGen is a fast context-loading module for LLM systems. First, CacheGen uses a custom tensor encoder, leveraging KV cache's distributional properties to encode a KV cache into more compact bitstream representations with negligible decoding overhead, to save bandwidth usage. Second, CacheGen adapts the compression level of different parts of a KV cache to cope with changes in available bandwidth, in order to maintain low context-loading delay and high generation quality. % When available bandwidth drops, CacheGen may raise the compression level for a part of the context or recompute its KV cache on the fly. We test CacheGen on popular LLMs and datasets. Compared to the recent systems that reuse the KV cache, CacheGen reduces the KV cache size by 3.5-4.3x and the total delay in fetching and processing contexts by 3.2-3.7x with negligible impact on the LLM response quality. Our code is at: https://github.com/UChi-JCL/CacheGen.

Yuhan Liu, Shangbin Feng, Xiaochuang Han et al. · cmu

In this work, we take a first step towards designing summarization systems that are faithful to the author's intent, not only the semantic content of the article. Focusing on a case study of preserving political perspectives in news summarization, we find that existing approaches alter the political opinions and stances of news articles in more than 50% of summaries, misrepresenting the intent and perspectives of the news authors. We thus propose P^3SUM, a diffusion model-based summarization approach controlled by political perspective classifiers. In P^3SUM, the political leaning of a generated summary is iteratively evaluated at each decoding step, and any drift from the article's original stance incurs a loss back-propagated to the embedding layers, steering the political stance of the summary at inference time. Extensive experiments on three news summarization datasets demonstrate that P^3SUM outperforms state-of-the-art summarization systems and large language models by up to 13.7% in terms of the success rate of stance preservation, with competitive performance on standard metrics of summarization quality. Our findings present a first analysis of preservation of pragmatic features in summarization, highlight the lacunae in existing summarization models -- that even state-of-the-art models often struggle to preserve author's intents -- and develop new summarization systems that are more faithful to author's perspectives.

Kuntai Du, Yuhan Liu, Yitian Hao et al. · stanford

Deep learning inference on streaming media data, such as object detection in video or LiDAR feeds and text extraction from audio waves, is now ubiquitous. To achieve high inference accuracy, these applications typically require significant network bandwidth to gather high-fidelity data and extensive GPU resources to run deep neural networks (DNNs). While the high demand for network bandwidth and GPU resources could be substantially reduced by optimally adapting the configuration knobs, such as video resolution and frame rate, current adaptation techniques fail to meet three requirements simultaneously: adapt configurations (i) with minimum extra GPU or bandwidth overhead; (ii) to reach near-optimal decisions based on how the data affects the final DNN's accuracy, and (iii) do so for a range of configuration knobs. This paper presents OneAdapt, which meets these requirements by leveraging a gradient-ascent strategy to adapt configuration knobs. The key idea is to embrace DNNs' differentiability to quickly estimate the accuracy's gradient to each configuration knob, called AccGrad. Specifically, OneAdapt estimates AccGrad by multiplying two gradients: InputGrad (i.e. how each configuration knob affects the input to the DNN) and DNNGrad (i.e. how the DNN input affects the DNN inference output). We evaluate OneAdapt across five types of configurations, four analytic tasks, and five types of input data. Compared to state-of-the-art adaptation schemes, OneAdapt cuts bandwidth usage and GPU usage by 15-59% while maintaining comparable accuracy or improves accuracy by 1-5% while using equal or fewer resources.

Wenxuan Ding, Shangbin Feng, Yuhan Liu et al. · cmu

We propose Knowledge Crosswords, a geometric knowledge reasoning benchmark consisting of incomplete knowledge networks bounded by structured factual constraints, where LLMs are tasked with inferring the missing facts to meet all constraints. The novel setting of geometric knowledge reasoning necessitates new LM abilities beyond existing atomic/linear multi-hop QA, such as backtracking, verifying facts and constraints, reasoning with uncertainty, and more. Knowledge Crosswords contains 2,101 individual problems, covering diverse knowledge domains, and is further divided into three difficulty levels. We conduct extensive experiments to evaluate existing LLMs and approaches on Knowledge Crosswords. Results demonstrate that baseline approaches struggle with larger knowledge networks and semantically-equivalent entity distractors. In light of their limitations, we propose two new approaches, Staged Prompting and Verify-All, to augment LLMs' abilities for error-aware backtracking and constraint verification. Our Verify-All significantly outperforms prior methods and is more robust towards problems in the hard subset. Further analysis shows that geometric knowledge reasoning poses new challenges to LLMs' knowledge abilities, particularly in robustness towards varying option orders, complex structural constraints in knowledge networks, "none of the above" scenarios, and more.

Shaoting Feng, Yuhan Liu, Hanchen Li et al. · stanford

Reusing KV cache is essential for high efficiency of Large Language Model (LLM) inference systems. With more LLM users, the KV cache footprint can easily exceed GPU memory capacity, so prior work has proposed to either evict KV cache to lower-tier storage devices, or compress KV cache so that more KV cache can be fit in the fast memory. However, prior work misses an important opportunity: jointly optimizing the eviction and compression decisions across all KV caches to minimize average generation latency without hurting quality. We propose EVICPRESS, a KV-cache management system that applies lossy compression and adaptive eviction to KV cache across multiple storage tiers. Specifically, for each KV cache of a context, EVICPRESS considers the effect of compression and eviction of the KV cache on the average generation quality and delay across all contexts as a whole. To achieve this, EVICPRESS proposes a unified utility function that quantifies the effect of quality and delay of the lossy compression or eviction. To this end, EVICPRESS's profiling module periodically updates the utility function scores on all possible eviction-compression configurations for all contexts and places KV caches using a fast heuristic to rearrange KV caches on all storage tiers, with the goal of maximizing the utility function scores on each storage tier. Compared to the baselines that evict KV cache or compress KV cache, EVICPRESS achieves higher KV-cache hit rates on fast devices, i.e., lower delay, while preserving high generation quality by applying conservative compression to contexts that are sensitive to compression errors. Evaluation on 12 datasets and 5 models demonstrates that EVICPRESS achieves up to 2.19x faster time-to-first-token (TTFT) at equivalent generation quality.

Shangbin Feng, Zhaoxuan Tan, Herun Wan et al.

Twitter bot detection has become an increasingly important task to combat misinformation, facilitate social media moderation, and preserve the integrity of the online discourse. State-of-the-art bot detection methods generally leverage the graph structure of the Twitter network, and they exhibit promising performance when confronting novel Twitter bots that traditional methods fail to detect. However, very few of the existing Twitter bot detection datasets are graph-based, and even these few graph-based datasets suffer from limited dataset scale, incomplete graph structure, as well as low annotation quality. In fact, the lack of a large-scale graph-based Twitter bot detection benchmark that addresses these issues has seriously hindered the development and evaluation of novel graph-based bot detection approaches. In this paper, we propose TwiBot-22, a comprehensive graph-based Twitter bot detection benchmark that presents the largest dataset to date, provides diversified entities and relations on the Twitter network, and has considerably better annotation quality than existing datasets. In addition, we re-implement 35 representative Twitter bot detection baselines and evaluate them on 9 datasets, including TwiBot-22, to promote a fair comparison of model performance and a holistic understanding of research progress. To facilitate further research, we consolidate all implemented codes and datasets into the TwiBot-22 evaluation framework, where researchers could consistently evaluate new models and datasets. The TwiBot-22 Twitter bot detection benchmark and evaluation framework are publicly available at https://twibot22.github.io/

Yuhan Liu, Ananda Theertha Suresh, Wennan Zhu et al.

We study the problem of histogram estimation under user-level differential privacy, where the goal is to preserve the privacy of all entries of any single user. We consider the heterogeneous scenario where the quantity of data can be different for each user. In this scenario, the amount of noise injected into the histogram to obtain differential privacy is proportional to the maximum user contribution, which can be amplified by few outliers. One approach to circumvent this would be to bound (or limit) the contribution of each user to the histogram. However, if users are limited to small contributions, a significant amount of data will be discarded. In this work, we propose algorithms to choose the best user contribution bound for histogram estimation under both bounded and unbounded domain settings. When the size of the domain is bounded, we propose a user contribution bounding strategy that almost achieves a two-approximation with respect to the best contribution bound in hindsight. For unbounded domain histogram estimation, we propose an algorithm that is logarithmic-approximation with respect to the best contribution bound in hindsight. This result holds without any distribution assumptions on the data. Experiments on both real and synthetic datasets verify our theoretical findings and demonstrate the effectiveness of our algorithms. We also show that clipping bias introduced by bounding user contribution may be reduced under mild distribution assumptions, which can be of independent interest.

Yixuan Liu, Suyun Zhao, Li Xiong et al.

Federated Learning, as a popular paradigm for collaborative training, is vulnerable against privacy attacks. Different privacy levels regarding users' attitudes need to be satisfied locally, while a strict privacy guarantee for the global model is also required centrally. Personalized Local Differential Privacy (PLDP) is suitable for preserving users' varying local privacy, yet only provides a central privacy guarantee equivalent to the worst-case local privacy level. Thus, achieving strong central privacy as well as personalized local privacy with a utility-promising model is a challenging problem. In this work, a general framework (APES) is built up to strengthen model privacy under personalized local privacy by leveraging the privacy amplification effect of the shuffle model. To tighten the privacy bound, we quantify the heterogeneous contributions to the central privacy user by user. The contributions are characterized by the ability of generating "echos" from the perturbation of each user, which is carefully measured by proposed methods Neighbor Divergence and Clip-Laplace Mechanism. Furthermore, we propose a refined framework (S-APES) with the post-sparsification technique to reduce privacy loss in high-dimension scenarios. To the best of our knowledge, the impact of shuffling on personalized local privacy is considered for the first time. We provide a strong privacy amplification effect, and the bound is tighter than the baseline result based on existing methods for uniform local privacy. Experiments demonstrate that our frameworks ensure comparable or higher accuracy for the global model.

Jayadev Acharya, Yuhan Liu, Ziteng Sun

We study discrete distribution estimation under user-level local differential privacy (LDP). In user-level $\varepsilon$-LDP, each user has $m\ge1$ samples and the privacy of all $m$ samples must be preserved simultaneously. We resolve the following dilemma: While on the one hand having more samples per user should provide more information about the underlying distribution, on the other hand, guaranteeing the privacy of all $m$ samples should make the estimation task more difficult. We obtain tight bounds for this problem under almost all parameter regimes. Perhaps surprisingly, we show that in suitable parameter regimes, having $m$ samples per user is equivalent to having $m$ times more users, each with only one sample. Our results demonstrate interesting phase transitions for $m$ and the privacy parameter $\varepsilon$ in the estimation risk. Finally, connecting with recent results on shuffled DP, we show that combined with random shuffling, our algorithm leads to optimal error guarantees (up to logarithmic factors) under the central model of user-level DP in certain parameter regimes. We provide several simulations to verify our theoretical findings.

Pengzhi Yang, Yuhan Liu, Shumon Koga et al.

This paper proposes a method for learning continuous control policies for active landmark localization and exploration using an information-theoretic cost. We consider a mobile robot detecting landmarks within a limited sensing range, and tackle the problem of learning a control policy that maximizes the mutual information between the landmark states and the sensor observations. We employ a Kalman filter to convert the partially observable problem in the landmark state to Markov decision process (MDP), a differentiable field of view to shape the reward, and an attention-based neural network to represent the control policy. The approach is further unified with active volumetric mapping to promote exploration in addition to landmark localization. The performance is demonstrated in several simulated landmark localization tasks in comparison with benchmark methods.

Chengyi Yang, Zhishang Xiang, Yunbo Tang et al.

Test-Time Training offers a promising way to improve the reasoning ability of large language models (LLMs) by adapting the model using only the test questions. However, existing methods struggle with difficult reasoning problems for two reasons: raw test questions are often too difficult to yield high-quality pseudo-labels, and the limited size of test sets makes continuous online updates prone to instability. To address these limitations, we propose TTCS, a co-evolving test-time training framework. Specifically, TTCS initializes two policies from the same pretrained model: a question synthesizer and a reasoning solver. These policies evolve through iterative optimization: the synthesizer generates progressively challenging question variants conditioned on the test questions, creating a structured curriculum tailored to the solver's current capability, while the solver updates itself using self-consistency rewards computed from multiple sampled responses on both original test and synthetic questions. Crucially, the solver's feedback guides the synthesizer to generate questions aligned with the model's current capability, and the generated question variants in turn stabilize the solver's test-time training. Experiments show that TTCS consistently strengthens the reasoning ability on challenging mathematical benchmarks and transfers to general-domain tasks across different LLM backbones, highlighting a scalable path towards dynamically constructing test-time curricula for self-evolving. Our code and implementation details are available at https://github.com/XMUDeepLIT/TTCS.

Xinyu Zhang, Yuhan Liu, Yuting Wang et al.

Few-shot object detection aims at detecting novel categories given only a few example images. It is a basic skill for a robot to perform tasks in open environments. Recent methods focus on finetuning strategies, with complicated procedures that prohibit a wider application. In this paper, we introduce DE-ViT, a few-shot object detector without the need for finetuning. DE-ViT's novel architecture is based on a new region-propagation mechanism for localization. The propagated region masks are transformed into bounding boxes through a learnable spatial integral layer. Instead of training prototype classifiers, we propose to use prototypes to project ViT features into a subspace that is robust to overfitting on base classes. We evaluate DE-ViT on few-shot, and one-shot object detection benchmarks with Pascal VOC, COCO, and LVIS. DE-ViT establishes new state-of-the-art results on all benchmarks. Notably, for COCO, DE-ViT surpasses the few-shot SoTA by 15 mAP on 10-shot and 7.2 mAP on 30-shot and one-shot SoTA by 2.8 AP50. For LVIS, DE-ViT outperforms few-shot SoTA by 17 box APr. Further, we evaluate DE-ViT with a real robot by building a pick-and-place system for sorting novel objects based on example images. The videos of our robot demonstrations, the source code and the models of DE-ViT can be found at https://mlzxy.github.io/devit.

Gang Liu, Sobin Alosious, Subhamoy Mahajan et al.

Machine learning (ML) offers a powerful path toward discovering sustainable polymer materials, but progress has been limited by the lack of large, high-quality, and openly accessible polymer datasets. The Open Polymer Challenge (OPC) addresses this gap by releasing the first community-developed benchmark for polymer informatics, featuring a dataset with 10K polymers and 5 properties: thermal conductivity, radius of gyration, density, fractional free volume, and glass transition temperature. The challenge centers on multi-task polymer property prediction, a core step in virtual screening pipelines for materials discovery. Participants developed models under realistic constraints that include small data, label imbalance, and heterogeneous simulation sources, using techniques such as feature-based augmentation, transfer learning, self-supervised pretraining, and targeted ensemble strategies. The competition also revealed important lessons about data preparation, distribution shifts, and cross-group simulation consistency, informing best practices for future large-scale polymer datasets. The resulting models, analysis, and released data create a new foundation for molecular AI in polymer science and are expected to accelerate the development of sustainable and energy-efficient materials. Along with the competition, we release the test dataset at https://www.kaggle.com/datasets/alexliu99/neurips-open-polymer-prediction-2025-test-data. We also release the data generation pipeline at https://github.com/sobinalosious/ADEPT, which simulates more than 25 properties, including thermal conductivity, radius of gyration, and density.

Ruoxi Su, Yuhan Liu, Jingyu Hu

Accurately simulating the decisions of a specific individual remains challenging for large language models (LLMs), partly because persona information is often provided as static descriptions that miss the values, experiences, and contextual cues needed for individual-level decision simulation. We propose an adaptive interview framework that gathers persona-relevant information through a structured three-stage dialogue: core questions, dynamic follow-ups, and a synthesized personality summary. Using the resulting interview transcripts, we evaluate whether LLMs can simulate participants' decisions in moral dilemma scenarios. We compare three conversational contexts -- Core-10 responses, the full interview dialogue, and a summarized persona representation. We find that adaptive interviewing functions less as a uniform accuracy booster and more as a selective grounding mechanism: follow-up-derived evidence is incorporated in around 40% of full-interview traces, and these follow-up-grounded predictions are more accurate than core-only grounded ones (45.5% vs. 39.3%). These findings highlight that richer persona context alone is insufficient: improvements arise only when models actually ground their decisions in user-specific evidence.

Yuhan Liu, Anna Fang, Glen Moriarty et al.

Online mental health communities (OMHCs) are an effective and accessible channel to give and receive social support for individuals with mental and emotional issues. However, a key challenge on these platforms is finding suitable partners to interact with given that mechanisms to match users are currently underdeveloped. In this paper, we collaborate with one of the world's largest OMHC to develop an agent-based simulation framework and explore the trade-offs in different matching algorithms. The simulation framework allows us to compare current mechanisms and new algorithmic matching policies on the platform, and observe their differing effects on a variety of outcome metrics. Our findings include that usage of the deferred-acceptance algorithm can significantly better the experiences of support-seekers in one-on-one chats while maintaining low waiting time. We note key design considerations that agent-based modeling reveals in the OMHC context, including the potential benefits of algorithmic matching on marginalized communities.

Yuhan Liu, Jun Gao, Jiachen Du et al. · tencent-ai

A good empathetic dialogue system should first track and understand a user's emotion and then reply with an appropriate emotion. However, current approaches to this task either focus on improving the understanding of users' emotion or on proposing better responding strategies, and very few works consider both at the same time. Our work attempts to fill this vacancy. Inspired by task-oriented dialogue systems, we propose a novel empathetic response generation model with emotion-aware dialogue management. The emotion-aware dialogue management contains two parts: (1) Emotion state tracking maintains the current emotion state of the user and (2) Empathetic dialogue policy selection predicts a target emotion and a user's intent based on the results of the emotion state tracking. The predicted information is then used to guide the generation of responses. Experimental results show that dynamically managing different information can help the model generate more empathetic responses compared with several baselines under both automatic and human evaluations.

Yuhan Liu, Shahriar Negaharipour

We propose an optimization technique for 3-D underwater object modeling from 2-D forward-scan sonar images at known poses. A key contribution, for objects imaged in the proximity of the sea surface, is to resolve the multipath artifacts due to the air-water interface. Here, the object image formed by the direct target backscatter is almost always corrupted by the ghost and sometimes by the mirror components (generated by the multipath propagation). Assuming a planar air-water interface, we model, localize, and discard the corrupted object region within each view, thus avoiding the distortion of recovered 3-D shape. Additionally, complementary visual cues from the boundary of the mirror component, distinct at suitable sonar poses, are employed to enhance the 3-D modeling accuracy. The optimization is implemented as iterative shape adjustment by displacing the vertices of triangular patches in the 3-D surface mesh model, in order to minimize the discrepancy between the data and synthesized views of the 3-D object model. To this end, we first determine 2-D motion fields that align the object regions in the data and synthesized views, then calculate the 3-D motion of triangular patch centers, and finally the model vertices. The 3-D model is initialized with the solution of an earlier space carving method applied to the same data. The same parameters are applied in various experiments with 2 real data sets, mixed real-synthetic data set, and computer-generated data guided by general findings from a real experiment, to explore the impact of non-flat air-water interface. The results confirm the generation of a refined 3-D model in about half-dozen iterations.

Haonan Chang, Kowndinya Boyalakuntla, Yuhan Liu et al.

Solving storage problem: where objects must be accurately placed into containers with precise orientations and positions, presents a distinct challenge that extends beyond traditional rearrangement tasks. These challenges are primarily due to the need for fine-grained 6D manipulation and the inherent multi-modality of solution spaces, where multiple viable goal configurations exist for the same storage container. We present a novel Diffusion-based Affordance Prediction (DAP) pipeline for the multi-modal object storage problem. DAP leverages a two-step approach, initially identifying a placeable region on the container and then precisely computing the relative pose between the object and that region. Existing methods either struggle with multi-modality issues or computation-intensive training. Our experiments demonstrate DAP's superior performance and training efficiency over the current state-of-the-art RPDiff, achieving remarkable results on the RPDiff benchmark. Additionally, our experiments showcase DAP's data efficiency in real-world applications, an advancement over existing simulation-driven approaches. Our contribution fills a gap in robotic manipulation research by offering a solution that is both computationally efficient and capable of handling real-world variability. Code and supplementary material can be found at: https://github.com/changhaonan/DPS.git.

Haoqiang Kang, Xiaokang Ye, Yuhan Liu et al.

LLM/VLM-based digital agents have advanced rapidly thanks to scalable sandboxes for coding, web navigation, and computer use, which provide rich interactive training grounds. In contrast, embodied agents still lack abundant, diverse, and automatically generated 3D environments for interactive learning. Existing embodied simulators rely on manually crafted scenes or procedural templates, while recent LLM-based 3D generation systems mainly produce static scenes rather than deployable environments with verifiable tasks and standard learning interfaces. We introduce SimWorld Studio, an open-source platform built on Unreal Engine 5 for generating evolving embodied learning environments. At its core is SimCoder, a tool/skill-augmented coding agent that writes and executes engine-level code to construct physically grounded 3D worlds from language/image instructions. SimCoder self-evolves by using verifier feedback (e.g., compilation errors, physics checks, VLM critiques) to revise environments and autonomously add reusable tools and skills to its library. Generated worlds are exported as Gym-style environments for embodied agent learning. SimWorld Studio further enables co-evolution between environment generation and embodied learning: agent performance feedback guides SimCoder to generate adaptive curricula near the learner's capability frontier, so that environments become increasingly challenging as the embodied agent improves. Three case studies on embodied navigation show that self-evolution improves generation reliability, generated environments substantially improve embodied agent performance that generalizes to unseen benchmarks, and co-evolution yields an 18-point success-rate gain over fixed-environment learning and a 40-point gain over an untrained agent.

Xiantai Xiang, Guangyao Zhou, Zixiao Wen et al.

Multimodal object detection leveraging RGB and Infrared (IR) images is pivotal for robust perception in all-weather scenarios. While recent adapter-based approaches efficiently transfer RGB-pretrained foundation models to this task, they often prioritize model efficiency at the expense of cross-modal structural consistency. Consequently, critical structural cues are frequently lost when significant domain gaps arise, such as in high-contrast or nighttime environments. Moreover, conventional static multimodal fusion mechanisms typically lack environmental awareness, resulting in suboptimal adaptation and constrained detection performance under complex, dynamic scene variations. To address these limitations, we propose SLGNet, a parameter-efficient framework that synergizes hierarchical structural priors and language-guided modulation within a frozen Vision Transformer (ViT)-based foundation model. Specifically, we design a Structure-Aware Adapter to extract hierarchical structural representations from both modalities and dynamically inject them into the ViT to compensate for structural degradation inherent in ViT-based backbones. Furthermore, we propose a Language-Guided Modulation module that exploits VLM-driven structured captions to dynamically recalibrate visual features, thereby endowing the model with robust environmental awareness. Extensive experiments on the LLVIP, FLIR, KAIST, and DroneVehicle datasets demonstrate that SLGNet establishes new state-of-the-art performance. Notably, on the LLVIP benchmark, our method achieves an mAP of 66.1, while reducing trainable parameters by approximately 87% compared to traditional full fine-tuning. This confirms SLGNet as a robust and efficient solution for multimodal perception.

Yuhan Liu, Varun Nagaraj Rao, Sohyeon Hwang et al.

In this vision paper, we outline a blueprint for a decentralized network for the delivery industry, powered by an open protocol. By presenting the network's key components and layers, alongside hypothetical scenarios, we illustrate how the network and the protocol may function in practice. Through this decentralized approach, we aim to address three major issues that mark the current platform-based delivery economy: power imbalances between the platform and workers, information asymmetries caused by opaque decision-making, and value misalignments. Our goal is to provoke dialogue and inspire future work toward more equitable, transparent, and worker-centered futures in the delivery industry, the broader gig economy, and related domains.

Zirui Song, Bin Yan, Yuhan Liu et al.

Large Language Models (LLMs) have demonstrated remarkable success in various tasks such as natural language understanding, text summarization, and machine translation. However, their general-purpose nature often limits their effectiveness in domain-specific applications that require specialized knowledge, such as healthcare, chemistry, or legal analysis. To address this, researchers have explored diverse methods to enhance LLMs by integrating domain-specific knowledge. In this survey, we provide a comprehensive overview of these methods, which we categorize into four key approaches: dynamic knowledge injection, static knowledge embedding, modular adapters, and prompt optimization. Each approach offers unique mechanisms to equip LLMs with domain expertise, balancing trade-offs between flexibility, scalability, and efficiency. We discuss how these methods enable LLMs to tackle specialized tasks, compare their advantages and disadvantages, evaluate domain-specific LLMs against general LLMs, and highlight the challenges and opportunities in this emerging field. For those interested in delving deeper into this area, we also summarize the commonly used datasets and benchmarks. To keep researchers updated on the latest studies, we maintain an open-source at: https://github.com/abilliyb/Knowledge_Injection_Survey_Papers, dedicated to documenting research in the field of specialized LLM.

Yuqian Fu, Xingyu Qiu, Bin Ren et al.

Cross-Domain Few-Shot Object Detection (CD-FSOD) poses significant challenges to existing object detection and few-shot detection models when applied across domains. In conjunction with NTIRE 2025, we organized the 1st CD-FSOD Challenge, aiming to advance the performance of current object detectors on entirely novel target domains with only limited labeled data. The challenge attracted 152 registered participants, received submissions from 42 teams, and concluded with 13 teams making valid final submissions. Participants approached the task from diverse perspectives, proposing novel models that achieved new state-of-the-art (SOTA) results under both open-source and closed-source settings. In this report, we present an overview of the 1st NTIRE 2025 CD-FSOD Challenge, highlighting the proposed solutions and summarizing the results submitted by the participants.

Yanda Li, Yuhan Liu, Zirui Song et al.

Large audio-language models (LALMs) generalize across speech, sound, and music, but unified decoders can exhibit a \emph{temporal smoothing bias}: transient acoustic cues may be underutilized in favor of temporally smooth context that is better supported by language priors, leading to less specific audio-grounded outputs. We propose \emph{Temporal Contrastive Decoding} (TCD), a training-free decoding method for unified LALMs that mitigates this effect at inference time. TCD constructs a temporally blurred slow-path view by smoothing the input waveform and re-encoding it, then contrasts next-token logits from the original and slow-path views. The contrastive signal is applied as a token-level logit update restricted to a small candidate set. A self-normalized stability score sets the blur window and update scale, and a step-wise gate based on uncertainty and audio reliance activates the update only when needed. Experiments on MMAU and AIR-Bench show consistent improvements on strong unified LALMs. We further conduct ablations and an architectural applicability study to analyze the contributions of key components and how TCD behaves across large audio-language model designs.

Yubo Wang, Juntian Zhang, Yichen Wu et al.

While Chain-of-Thought empowers Large Vision-Language Models with multi-step reasoning, explicit textual rationales suffer from an information bandwidth bottleneck, where continuous visual details are discarded during discrete tokenization. Recent latent reasoning methods attempt to address this challenge, but often fall prey to premature semantic collapse due to rigid autoregressive objectives. In this paper, we propose Laser, a novel paradigm that reformulates visual deduction via Dynamic Windowed Alignment Learning (DWAL). Instead of forcing a point-wise prediction, Laser aligns the latent state with a dynamic validity window of future semantics. This mechanism enforces a "Forest-before-Trees" cognitive hierarchy, enabling the model to maintain a probabilistic superposition of global features before narrowing down to local details. Crucially, Laser maintains interpretability via decodable trajectories while stabilizing unconstrained learning via Self-Refined Superposition. Extensive experiments on 6 benchmarks demonstrate that Laser achieves state-of-the-art performance among latent reasoning methods, surpassing the strong baseline Monet by 5.03% on average. Notably, it achieves these gains with extreme efficiency, reducing inference tokens by more than 97%, while demonstrating robust generalization to out-of-distribution domains.

Huamin Chen, Xunzhuo Liu, Bowei He et al.

Over the past year, the vLLM Semantic Router project has released a series of work spanning: (1) core routing mechanisms -- signal-driven routing, context-length pool routing, router performance engineering, policy conflict detection, low-latency embedding models, category-aware semantic caching, user-feedback-driven routing adaptation, hallucination detection, and hierarchical content-safety classification for privacy and jailbreak protection; (2) fleet optimization -- fleet provisioning and energy-efficiency analysis; (3) agentic and multimodal routing -- multimodal agent routing, tool selection, CUA security, and multi-turn context memory and safety; (4) governance and standards -- inference routing protocols and multi-provider API extensions. Each paper tackled a specific problem in LLM inference, but the problems are not independent; for example, fleet provisioning depends on the routing policy, which depends on the workload mix, shifting as organizations adopt agentic and multimodal workloads. This paper distills those results into the Workload-Router-Pool (WRP) architecture, a three-dimensional framework for LLM inference optimization. Workload characterizes what the fleet serves (chat vs. agent, single-turn vs. multi-turn, warm vs. cold, prefill-heavy vs. decode-heavy). Router determines how each request is dispatched (static semantic rules, online bandit adaptation, RL-based model selection, quality-aware cascading). Pool defines where inference runs (homogeneous vs. heterogeneous GPU, disaggregated prefill/decode, KV-cache topology). We map our prior work onto a 3x3 WRP interaction matrix, identify which cells we have covered and which remain open, and propose twenty-one concrete research directions at the intersections, each grounded in our prior measurements, tiered by maturity from engineering-ready to open research.

Yuhan Liu, Chengcheng Wan, Kuntai Du et al.

ML APIs have greatly relieved application developers of the burden to design and train their own neural network models -- classifying objects in an image can now be as simple as one line of Python code to call an API. However, these APIs offer the same pre-trained models regardless of how their output is used by different applications. This can be suboptimal as not all ML inference errors can cause application failures, and the distinction between inference errors that can or cannot cause failures varies greatly across applications. To tackle this problem, we first study 77 real-world applications, which collectively use six ML APIs from two providers, to reveal common patterns of how ML API output affects applications' decision processes. Inspired by the findings, we propose ChameleonAPI, an optimization framework for ML APIs, which takes effect without changing the application source code. ChameleonAPI provides application developers with a parser that automatically analyzes the application to produce an abstract of its decision process, which is then used to devise an application-specific loss function that only penalizes API output errors critical to the application. ChameleonAPI uses the loss function to efficiently train a neural network model customized for each application and deploys it to serve API invocations from the respective application via existing interface. Compared to a baseline that selects the best-of-all commercial ML API, we show that ChameleonAPI reduces incorrect application decisions by 43%.

Huamin Chen, Xunzhuo Liu, Yuhan Liu et al.

Sizing a GPU fleet for LLM inference is harder than it looks. The obvious questions -- how many GPUs, which type, where to split a two-pool fleet -- have no closed-form answers. They depend on the full token-length distribution, the routing policy, and queueing dynamics that turn ugly under heavy-tailed workloads. Existing tools optimize per-engine configuration for a fixed GPU count; none of them address the upstream question of how many GPUs to buy and how to arrange them. inference-fleet-sim fills that gap. It combines analytical M/G/c queueing with discrete-event simulation (DES) to find the minimum-cost fleet configuration that empirically meets a P99 TTFT SLO. It includes a physics-informed GPU performance model covering A10G, A100, and H100 across monolithic, two-pool-routed, and disaggregated topologies, all without requiring access to real hardware. We run the tool on seven fleet-planning scenarios drawn from two public workload traces (LMSYS, Azure) and one synthetic agent-heavy trace. Each one surfaces a result that simple analysis gets wrong -- the right split threshold, the cheapest GPU type, whether an apparently idle fleet is actually broken -- and shows why joint simulation of queueing, routing, and hardware is necessary to find it.

Xiaoqing Zhang, Ang Lv, Yuhan Liu et al.

Large language models (LLMs) excel at few-shot in-context learning (ICL) without requiring parameter updates. However, as ICL demonstrations increase from a few to many, performance tends to plateau and eventually decline. We identify two primary causes for this trend: the suboptimal negative log-likelihood (NLL) optimization objective and the incremental data noise. To address these issues, we introduce \textit{DrICL}, a novel optimization method that enhances model performance through \textit{Differentiated} and \textit{Reweighting} objectives. Globally, DrICL utilizes differentiated learning to optimize the NLL objective, ensuring that many-shot performance surpasses zero-shot levels. Locally, it dynamically adjusts the weighting of many-shot demonstrations by leveraging cumulative advantages inspired by reinforcement learning, thereby mitigating the impact of noisy data. Recognizing the lack of multi-task datasets with diverse many-shot distributions, we develop the \textit{Many-Shot ICL Benchmark} (ICL-50)-a large-scale benchmark of 50 tasks that cover shot numbers from 1 to 350 within sequences of up to 8,000 tokens-for both fine-tuning and evaluation purposes. Experimental results demonstrate that LLMs enhanced with DrICL achieve significant improvements in many-shot setups across various tasks, including both in-domain and out-of-domain scenarios. We release the code and dataset hoping to facilitate further research in many-shot ICL\footnote{https://github.com/xiaoqzhwhu/DrICL}.

Jiahao Qiu, Yinghui He, Xinzhe Juan et al.

The rise of LLM-driven AI characters raises safety concerns, particularly for vulnerable human users with psychological disorders. To address these risks, we propose EmoAgent, a multi-agent AI framework designed to evaluate and mitigate mental health hazards in human-AI interactions. EmoAgent comprises two components: EmoEval simulates virtual users, including those portraying mentally vulnerable individuals, to assess mental health changes before and after interactions with AI characters. It uses clinically proven psychological and psychiatric assessment tools (PHQ-9, PDI, PANSS) to evaluate mental risks induced by LLM. EmoGuard serves as an intermediary, monitoring users' mental status, predicting potential harm, and providing corrective feedback to mitigate risks. Experiments conducted in popular character-based chatbots show that emotionally engaging dialogues can lead to psychological deterioration in vulnerable users, with mental state deterioration in more than 34.4% of the simulations. EmoGuard significantly reduces these deterioration rates, underscoring its role in ensuring safer AI-human interactions. Our code is available at: https://github.com/1akaman/EmoAgent

Owen Xingjian Zhang, Shuyao Zhou, Jiayi Geng et al.

In response to the increasing mental health challenges faced by college students, we sought to understand their perspectives on how AI applications, particularly Large Language Models (LLMs), can be leveraged to enhance their mental well-being. Through pilot interviews with ten diverse students, we explored their opinions on the use of LLMs across five fictional scenarios: General Information Inquiry, Initial Screening, Reshaping Patient-Expert Dynamics, Long-term Care, and Follow-up Care. Our findings revealed that students' acceptance of LLMs varied by scenario, with participants highlighting both potential benefits, such as proactive engagement and personalized follow-up care, and concerns, including limitations in training data and emotional support. These insights inform how AI technology should be designed and implemented to effectively support and enhance students' mental well-being, particularly in scenarios where LLMs can complement traditional methods, while maintaining empathy and respecting individual preferences.

Huamin Chen, Xunzhuo Liu, Yuhan Liu et al.

Modern LLM GPU fleets are provisioned for worst-case context lengths that the vast majority of requests never approach, wasting GPU capacity on idle KV-cache slots. We present FleetOpt, a framework that starts from first principles: given a workload's prompt-length CDF and a P99 TTFT target, derive the minimum-cost fleet analytically, then deploy it in practice. The analytical core models each pool as an M/G/c queue and derives that the minimum-cost fleet is a two-pool architecture -- a short-context pool and a long-context pool -- with an optimal boundary B* satisfying an equal marginal GPU cost condition across both pools. The fundamental barrier to achieving B* is the cost cliff: a hard routing step where requests just above B* consume 8x--42x more GPU capacity than requests just below it (depending on the context window ratio), creating a structural disincentive to lower the boundary. Compress-and-Route (C&R) is the implementation mechanism that resolves this barrier. Gateway-layer extractive compression trims borderline requests below B* before the engine ever sees them, converting the hard hardware boundary into a software parameter read from the workload CDF. The two components are unified in the FleetOpt offline planner: given a CDF and SLO, it returns the optimal (n_s*, n_l*, B*, gamma*) in under 1 ms. On three production traces, the combined framework reduces total GPU cost by 6--82% versus a homogeneous fleet, with C&R contributing 1--44 percentage points beyond plain pool routing depending on workload archetype. The analytical model is validated against a discrete-event simulator (inference-fleet-sim) with <= 3% error on predicted GPU utilization across all pools and workloads.

Yihua Cheng, Yuhan Liu, Jiayi Yao et al.

Today's LLM inference systems treat individual engines and queries independently for simplicity, but this causes significant resource inefficiencies. While there are proposals to avoid redundant computation by reusing KV caches across queries and to increase GPU utilization by disaggregating a single query to different engines, their promises cannot be realized without efficiently offloading and communicating KV cache across LLM inference engines and queries. We present LMCache, the first and so far the most efficient open-source KV caching solution, which extracts and stores KV caches generated by modern LLM engines (vLLM and SGLang) and shares the KV caches across engines and queries. LMCache exposes KV caches in the LLM engine interface, effectively transforming LLM engines from individual token processors to a collection of engines with KV cache as the storage and communication medium. In particular, it supports both cache offloading (prefix reuse across queries) and prefill-decode disaggregation (cross-engine cache transfer). LMCache's high performance and wide adoption stem from the following contributions: highly optimized KV cache data movement with performance optimizations including batched data movement operations, compute and I/O pipelining; a modular KV cache connector component, decoupling LMCache from the rapid evolution of inference engines; a first-class control API, such as pinning, lookup, cleanup, movement, and compression, for flexible cache orchestration across GPU, CPU, storage, and network layers. Evaluation shows that combining LMCache with vLLM achieves up to 15x improvement in throughput across diverse workloads. With a growing community, LMCache has seen dramatic growth in adoption by enterprise inference systems, which provides valuable lessons for future KV caching solutions. The source code of LMCache is at: https://github.com/LMCache/LMCache.

Yurun Chen, Xavier Hu, Yuhan Liu et al.

Large language models enable agents to autonomously perform tasks in open web environments. However, as hidden threats within the web evolve, web agents face the challenge of balancing task performance with emerging risks during long-sequence operations. Although this challenge is critical, current research remains limited to single-objective optimization or single-turn scenarios, lacking the capability for collaborative optimization of both safety and utility in web environments. To address this gap, we propose HarmonyGuard, a multi-agent collaborative framework that leverages policy enhancement and objective optimization to jointly improve both utility and safety. HarmonyGuard features a multi-agent architecture characterized by two fundamental capabilities: (1) Adaptive Policy Enhancement: We introduce the Policy Agent within HarmonyGuard, which automatically extracts and maintains structured security policies from unstructured external documents, while continuously updating policies in response to evolving threats. (2) Dual-Objective Optimization: Based on the dual objectives of safety and utility, the Utility Agent integrated within HarmonyGuard performs the Markovian real-time reasoning to evaluate the objectives and utilizes metacognitive capabilities for their optimization. Extensive evaluations on multiple benchmarks show that HarmonyGuard improves policy compliance by up to 38% and task completion by up to 20% over existing baselines, while achieving over 90% policy compliance across all tasks. Our project is available here: https://github.com/YurunChen/HarmonyGuard.

Jiayi Yao, Samuel Shen, Kuntai Du et al.

The large size of the KV cache has become a major bottleneck for serving LLMs with increasing context lengths. In response, many KV cache compression methods, such as token dropping and quantization, have been proposed. However, almost all of these methods are inherently lossy-despite minimal accuracy degradation for short outputs, their outputs increasingly diverge from full-KV-cache outputs as more tokens are decoded, which leads to catastrophic failures in code generation and tool calling. We present VeriCache, the first inference framework that ensures the same output as full-KV-cache decoding but largely preserves the high decoding throughput of a range of KV cache compression algorithms. VeriCache uses the compressed KV cache to draft tokens, then verifies them against the full KV cache. While it may seem like just speculative decoding, VeriCache requires addressing a key system challenge to work-keeping the full KV cache out of GPU memory and minimizing the overhead of swapping it in for verification. The insight is two-fold: (1) compressed-KV decoding can be parallelized with full-KV swap, because one is HBM-bandwidth-bound and the other is PCIe/network-bound, and (2) the compressed KV cache often produces output similar to the full KV cache, allowing a long drafting horizon to amortize each full-KV swap. VeriCache applies to both long-context decoding and remote prefix caching, supports a broad family of token-dropping and quantization methods through a uniform compressor interface, and composes with traditional speculative decoding. Experimental results show that VeriCache achieves up to 4X higher throughput than full-KV inference while producing identical outputs.

Yunbo Long, Yuhan Liu, Alexandra Brintrup

The deployment of large language models (LLMs) in automated negotiation has set a high performance benchmark, but their computational cost and data privacy requirements render them unsuitable for many privacy-sensitive, on-device applications such as mobile assistants, embodied AI agents or private client interactions. While small language models (SLMs) offer a practical alternative, they suffer from a significant performance gap compared to LLMs in playing emotionally charged complex personas, especially for credit negotiation. This paper introduces EQ-Negotiator, a novel framework that bridges this capability gap using emotional personas. Its core is a reasoning system that integrates game theory with a Hidden Markov Model(HMM) to learn and track debtor emotional states online, without pre-training. This allows EQ-Negotiator to equip SLMs with the strategic intelligence to counter manipulation while de-escalating conflict and upholding ethical standards. Through extensive agent-to-agent simulations across diverse credit negotiation scenarios, including adversarial debtor strategies like cheating, threatening, and playing the victim, we show that a 7B parameter language model with EQ-Negotiator achieves better debt recovery and negotiation efficiency than baseline LLMs more than 10 times its size. This work advances persona modeling from descriptive character profiles to dynamic emotional architectures that operate within privacy constraints. Besides, this paper establishes that strategic emotional intelligence, not raw model scale, is the critical factor for success in automated negotiation, paving the way for effective, ethical, and privacy-preserving AI negotiators that can operate on the edge.

Zhaolin Zhang, Mingqi Song, Wugang Meng et al.

In recent years, the millimeter-wave radar to identify human behavior has been widely used in medical,security, and other fields. When multiple radars are performing detection tasks, the validity of the features contained in each radar is difficult to guarantee. In addition, processing multiple radar data also requires a lot of time and computational cost. The Complementary Ensemble Empirical Mode Decomposition-Energy Slice (CEEMD-ES) multistatic radar selection method is proposed to solve these problems. First, this method decomposes and reconstructs the radar signal according to the difference in the reflected echo frequency between the limbs and the trunk of the human body. Then, the radar is selected according to the difference between the ratio of echo energy of limbs and trunk and the theoretical value. The time domain, frequency domain and various entropy features of the selected radar are extracted. Finally, the Extreme Learning Machine (ELM) recognition model of the ReLu core is established. Experiments show that this method can effectively select the radar, and the recognition rate of three kinds of human actions is 98.53%.

Zhen Liu, Yuhan Liu, Jinjun Wang et al.

Vision-and-Language Navigation requires agents to follow natural-language instructions in visually changing environments. A central challenge is the dynamic entanglement between language and observations: the meaning of instruction shifts as the agent's field of view and spatial context evolve. However, many existing models encode the instruction as a static global representation, limiting their ability to adapt instruction meaning to the current visual context. We therefore model instruction understanding as an Instruction-as-State variable: a decision-relevant, token-level instruction state that evolves step by step conditioned on the agent's perceptual state, where the perceptual state denotes the observation-grounded navigation context at each step. To realize this principle, we introduce State-Entangled Environment-Guided Instruction Understanding (S-EGIU), a coarse-to-fine framework for state-conditioned segment activation and token-level semantic refinement. At the coarse level, S-EGIU activates the instruction segment whose semantics align with the current observation. At the fine level, it refines the activated segment through observation-guided token grounding and contextual modeling, sharpening its internal semantics under the current observation. Together, these stages maintain an instruction state that is continuously updated according to the agent's perceptual state during navigation. S-EGIU delivers strong performance on several key metrics, including a +2.68% SPL gain on REVERIE Test Unseen, and demonstrates consistent efficiency gains across multiple VLN benchmarks, underscoring the value of dynamic instruction--perception entanglement.

Yukun Qi, Pei Fu, Hang Li et al.

Vision-Language Models (VLMs) have achieved remarkable progress on a wide range of challenging multimodal understanding and reasoning tasks. However, existing reasoning paradigms, such as the classical Chain-of-Thought (CoT), rely solely on textual information and often underutilize important visual cues. While prior work has incorporated pixel-level visual cues, these representations require precise spatial localization, introducing additional learning complexity. To address this, we propose PatchCue, a novel patch-based visual cue paradigm designed to significantly enhance the visual reasoning capabilities of VLMs. By partitioning images into patches and representing cues at the patch level, PatchCue aligns better with human perceptual habits and leverages the patch-tokenized input of modern VLMs. We train VLMs using a two-stage approach: cold-start supervised fine-tuning to output patch-level cues, followed by reinforcement learning with a process-supervised cue reward that guides intermediate visual reasoning steps. Extensive experiments on multiple VLMs and diverse benchmarks, including general visual question answering, complex reasoning, and document understanding, demonstrate that PatchCue consistently improves overall model performance. Our results show that patch-level cues outperform both pixel-level bounding boxes and point-based cues, providing a more effective and cognitively aligned visual reasoning paradigm.

Zinan Zeng, Sen Ye, Zijian Cai et al.

Online movie review websites are valuable for information and discussion about movies. However, the massive spoiler reviews detract from the movie-watching experience, making spoiler detection an important task. Previous methods simply focus on reviews' text content, ignoring the heterogeneity of information in the platform. For instance, the metadata and the corresponding user's information of a review could be helpful. Besides, the spoiler language of movie reviews tends to be genre-specific, thus posing a domain generalization challenge for existing methods. To this end, we propose MMoE, a multi-modal network that utilizes information from multiple modalities to facilitate robust spoiler detection and adopts Mixture-of-Experts to enhance domain generalization. MMoE first extracts graph, text, and meta feature from the user-movie network, the review's textual content, and the review's metadata respectively. To handle genre-specific spoilers, we then adopt Mixture-of-Experts architecture to process information in three modalities to promote robustness. Finally, we use an expert fusion layer to integrate the features from different perspectives and make predictions based on the fused embedding. Experiments demonstrate that MMoE achieves state-of-the-art performance on two widely-used spoiler detection datasets, surpassing previous SOTA methods by 2.56% and 8.41% in terms of accuracy and F1-score. Further experiments also demonstrate MMoE's superiority in robustness and generalization. Our code is available at https://github.com/zzqbjt/Spoiler-Detection.

Maryam Aliakbarpour, Vladimir Braverman, Yuhan Liu et al.

We study high-dimensional mean estimation in a collaborative setting where data is contributed by $N$ users in batches of size $n$. In this environment, a learner seeks to recover the mean $μ$ of a true distribution $P$ from a collection of sources that are both statistically heterogeneous and potentially malicious. We formalize this challenge through a double corruption landscape: an $\varepsilon$-fraction of users are entirely adversarial, while the remaining ``good'' users provide data from distributions that are related to $P$, but deviate by a proximity parameter $α$. Unlike existing work on the untrusted batch model, which typically measures this deviation via total variation distance in discrete settings, we address the continuous, high-dimensional regime under two natural variants for deviation: (1) good batches are drawn from distributions with a mean-shift of $\sqrtα$, or (2) an $α$-fraction of samples within each good batch are adversarially corrupted. In particular, the second model presents significant new challenges: in high dimensions, unlike discrete settings, even a small fraction of sample-level corruption can shift empirical means and covariances arbitrarily. We provide two Sum-of-Squares (SoS) based algorithms to navigate this tiered corruption. Our algorithms achieve the minimax-optimal error rate $O(\sqrt{\varepsilon/n} + \sqrt{d/nN} + \sqrtα)$, demonstrating that while heterogeneity $α$ represents an inherent statistical difficulty, the influence of adversarial users is suppressed by a factor of $1/\sqrt{n}$ due to the internal averaging afforded by the batch structure.

Yunbo Long, Yuhan Liu, Liming Xu et al.

The emergence of autonomous Large Language Model (LLM) agents has created a new ecosystem of strategic, agent-to-agent interactions. However, a critical challenge remains unaddressed: in high-stakes, emotion-sensitive domains like debt collection, LLM agents pre-trained on human dialogue are vulnerable to exploitation by adversarial counterparts who simulate negative emotions to derail negotiations. To fill this gap, we first contribute a novel dataset of simulated debt recovery scenarios and a multi-agent simulation framework. Within this framework, we introduce EmoDebt, an LLM agent architected for robust performance. Its core innovation is a Bayesian-optimized emotional intelligence engine that reframes a model's ability to express emotion in negotiation as a sequential decision-making problem. Through online learning, this engine continuously tunes EmoDebt's emotional transition policies, discovering optimal counter-strategies against specific debtor tactics. Extensive experiments on our proposed benchmark demonstrate that EmoDebt achieves significant strategic robustness, substantially outperforming non-adaptive and emotion-agnostic baselines across key performance metrics, including success rate and operational efficiency. By introducing both a critical benchmark and a robustly adaptive agent, this work establishes a new foundation for deploying strategically robust LLM agents in adversarial, emotion-sensitive debt interactions. The code is available at \textcolor{blue}{https://github.com/Yunbo-max/EmoDebt}.

Yihan Zhu, Yuhan Liu, Weijiang Li et al.

Despite the success of foundation models in language and vision, molecular graph generation still lacks a unified framework for heterogeneous design tasks with reliable controllability. While reinforcement learning (RL) offers a natural post-training mechanism for task-specific optimization, applying it to graph generative models is hindered by the vast atom-wise action spaces and chemically invalid intermediate states. We propose \textbf{Co}ntrollable \textbf{Mole}cular Generative Foundation Models (CoMole), built with a unified motif-aware graph diffusion pipeline. By learning a motif-aware graph space, CoMole transfers pretrained structural priors into controllable generation, where RL optimizes conditional reverse policies over chemically meaningful decisions. We theoretically characterize the bottleneck of atom-level RL and justify motif-aware policy optimization. Across three heterogeneous benchmarks spanning materials and drug discovery, CoMole ranks first in controllability on all nine targets, reduces MAE by up to 48.2% relative to the strongest baselines, and maintains validity above 0.94 without rule-based correction or post-hoc filtering. We further show that CoMole transfers controllability to unseen properties by optimizing only task embeddings with the generator frozen, achieving performance competitive with strong task-specific baselines.

Chong Tian, Yu Wang, Chenxu Yang et al.

Short-form video platforms are major channels for news but also fertile ground for multimodal misinformation where each modality appears plausible alone yet cross-modal relationships are subtly inconsistent, like mismatched visuals and captions. On two benchmark datasets, FakeSV (Chinese) and FakeTT (English), we observe a clear asymmetry: real videos exhibit high text-visual but moderate text-audio consistency, while fake videos show the opposite pattern. Moreover, a single global consistency score forms an interpretable axis along which fake probability and prediction errors vary smoothly. Motivated by these observations, we present MAGIC3 (Modal-Adversarial Gated Interaction and Consistency-Centric Classifier), a detector that explicitly models and exposes cross-tri-modal consistency signals at multiple granularities. MAGIC3 combines explicit pairwise and global consistency modeling with token- and frame-level consistency signals derived from cross-modal attention, incorporates multi-style LLM rewrites to obtain style-robust text representations, and employs an uncertainty-aware classifier for selective VLM routing. Using pre-extracted features, MAGIC3 consistently outperforms the strongest non-VLM baselines on FakeSV and FakeTT. While matching VLM-level accuracy, the two-stage system achieves 18-27x higher throughput and 93% VRAM savings, offering a strong cost-performance tradeoff.

Song Jin, Juntian Zhang, Yuhan Liu et al.

Evaluating and iterating upon recommender systems is crucial, yet traditional A/B testing is resource-intensive, and offline methods struggle with dynamic user-platform interactions. While agent-based simulation is promising, existing platforms often lack a mechanism for user actions to dynamically reshape the environment. To bridge this gap, we introduce RecInter, a novel agent-based simulation platform for recommender systems featuring a robust interaction mechanism. In RecInter platform, simulated user actions (e.g., likes, reviews, purchases) dynamically update item attributes in real-time, and introduced Merchant Agents can reply, fostering a more realistic and evolving ecosystem. High-fidelity simulation is ensured through Multidimensional User Profiling module, Advanced Agent Architecture, and LLM fine-tuned on Chain-of-Thought (CoT) enriched interaction data. Our platform achieves significantly improved simulation credibility and successfully replicates emergent phenomena like Brand Loyalty and the Matthew Effect. Experiments demonstrate that this interaction mechanism is pivotal for simulating realistic system evolution, establishing our platform as a credible testbed for recommender systems research. Our codes are available at https://github.com/jinsong8/RecInter.

Abtin Mahyar, Masoumeh Shafieinejad, Yuhan Liu et al.

Diffusion models are the leading approach for tabular data synthesis and are increasingly used to share sensitive records. Whether they actually protect privacy has become a pressing question. Membership inference attacks are the standard tool for this purpose, yet existing attacks assume a single-table setting and ignore the multi-relational structure of real sensitive data. A core challenge in assessing privacy risks from membership inference attacks in multi-table settings is how to leverage auxiliary information from relations associated with the target table, such as its parent tables. Particularly, we study a practical setting in which such auxiliary information is available only when training the attack model. At inference time, the attacker observes only the attribute values of the target record from the target table. We propose FERMI (FEature-mapping for Relational Membership Inference), which resolves this gap by enriching single-table features with relational membership signal. Across three tabular diffusion architectures and three real-world relational datasets, FERMI consistently improves attack performance over single-table baselines, with TPR@$0.1$FPR rising by up to 53% over the single-table baseline in the white-box setting and 22% in the black-box setting.

Zixiao Wen, Zhen Yang, Jiawei Li et al.

Single object tracking in satellite videos is inherently challenged by small target, blurred background, large aspect ratio changes, and frequent visual occlusions. These constraints often cause appearance-based trackers to accumulate errors and lose targets irreversibly. To systematically mitigate both spatial ambiguities and temporal information loss, we propose SiamGM, a novel geometry-aware and motion-guided Siamese network. From a spatial perspective, we introduce an Inter-Frame Graph Attention (IFGA) module, closely integrated with an Aspect Ratio-Constrained Label Assignment (LA) method, establishing fine-grained topological correspondences and explicitly preventing surrounding background noise. From a temporal perspective, we introduce the Motion Vector-Guided Online Tracking Optimization method. By adopting the Normalized Peak-to-Sidelobe Ratio (nPSR) as a dynamic confidence indicator, we propose an Online Motion Model Refinement (OMMR) strategy to utilize historical trajectory information. Evaluations on two challenging SatSOT and SV248S benchmarks confirm that SiamGM outperforms most state-of-the-art trackers in both precision and success metrics. Notably, the proposed components of SiamGM introduce virtually no computational overhead, enabling real-time tracking at 130 frames per second (FPS). Codes and tracking results are available at https://github.com/wenzx18/SiamGM.

Yuhan Liu, Sheng Wang, Yixuan Liu et al.

The Sparse Vector Technique (SVT) is one of the most fundamental tools in differential privacy (DP). It works as a backbone for adaptive data analysis by answering a sequence of queries on a given dataset, and gleaning useful information in a privacy-preserving manner. Unlike the typical private query releases that directly publicize the noisy query results, SVT is less informative -- it keeps the noisy query results to itself and only reveals a binary bit for each query, indicating whether the query result surpasses a predefined threshold. To provide a rigorous DP guarantee for SVT, prior works in the literature adopt a conservative privacy analysis by assuming the direct disclosure of noisy query results as in typical private query releases. This approach, however, hinders SVT from achieving higher query accuracy due to an overestimation of the privacy risks, which further leads to an excessive noise injection using the Laplacian or Gaussian noise for perturbation. Motivated by this, we provide a new privacy analysis for SVT by considering its less informative nature. Our analysis results not only broaden the range of applicable noise types for perturbation in SVT, but also identify the exponential noise as optimal among all evaluated noises (which, however, is usually deemed non-applicable in prior works). The main challenge in applying exponential noise to SVT is mitigating the sub-optimal performance due to the bias introduced by noise distributions. To address this, we develop a utility-oriented optimal threshold correction method and an appending strategy, which enhances the performance of SVT by increasing the precision and recall, respectively. The effectiveness of our proposed methods is substantiated both theoretically and empirically, demonstrating significant improvements up to $50\%$ across evaluated metrics.

Yuhan Liu, Lianhui Qin, Shengjie Wang

Large Vision-Language Models (VLMs) have achieved remarkable progress in multimodal understanding, yet they struggle when reasoning over information-intensive images that densely interleave textual annotations with fine-grained graphical elements. The main challenges lie in precisely localizing critical cues in dense layouts and multi-hop reasoning to integrate dispersed evidence. We propose Speculative Verdict (SV), a training-free framework inspired by speculative decoding that combines multiple lightweight draft experts with a large verdict model. In the draft stage, small VLMs act as draft experts to generate reasoning paths that provide diverse localization candidates; in the verdict stage, a strong VLM synthesizes these paths to produce the final answer, minimizing computational cost while recovering correct answers. To further improve efficiency and accuracy, SV introduces a consensus expert selection mechanism that forwards only high-agreement reasoning paths to the verdict. Empirically, SV achieves consistent gains on challenging information-intensive and high-resolution visual question answering benchmarks, including InfographicVQA, ChartMuseum, ChartQAPro, and HR-Bench 4K. By synthesizing correct insights from multiple partially accurate reasoning paths, SV achieves both error correction and cost-efficiency compared to large proprietary models or training pipelines. Code is available at https://github.com/Tinaliu0123/speculative-verdict

Chen Wang, Xunzhuo Liu, Yuhan Liu et al.

Large Language Models (LLMs) demonstrate substantial accuracy gains when augmented with reasoning modes such as chain-of-thought and inference-time scaling. However, reasoning also incurs significant costs in inference latency and token usage, with environmental and financial impacts, which are unnecessary for many simple prompts. We present a semantic router that classifies queries based on their reasoning requirements and selectively applies reasoning only when beneficial. Our approach achieves a 10.2 percentage point improvement in accuracy on the MMLU-Pro benchmark while reducing response latency by 47.1% and token consumption by 48.5% compared to direct inference with vLLM. These results demonstrate that semantic routing offers an effective mechanism for striking a balance between accuracy and efficiency in open-source LLM serving systems