Jiaqi Zheng

Shiju Zhao, Junhao Hu, Jiaqi Zheng et al.

The Key-Value (KV) cache of Large Language Models (LLMs) is prefix-based, making it highly inefficient for processing contexts retrieved in arbitrary order. Position-Independent Caching (PIC) has been proposed to enable KV reuse without positional constraints; however, existing approaches often incur substantial accuracy degradation, limiting their practical adoption. To address this issue, we propose native PIC by reintroducing the encoder to prevalent decoder-only LLMs and explicitly training it to support PIC. We further develop COMB, a PIC-aware caching system that integrates seamlessly with existing inference frameworks. Experimental results show that COMB reduces Time-to-First-Token (TTFT) by 51-94% and increases throughput by 3$\times$ with comparable accuracy. Furthermore, the quality improvement when using DeepSeek-V2-Lite-Chat demonstrates the applicability of COMB to other types of decoder-only LLMs. Our code is available at https://github.com/shijuzhao/Comb.

Hao Zhou, Shaoming Li, Guibin Jiang et al.

Marketing is an important mechanism to increase user engagement and improve platform revenue, and heterogeneous causal learning can help develop more effective strategies. Most decision-making problems in marketing can be formulated as resource allocation problems and have been studied for decades. Existing works usually divide the solution procedure into two fully decoupled stages, i.e., machine learning (ML) and operation research (OR) -- the first stage predicts the model parameters and they are fed to the optimization in the second stage. However, the error of the predicted parameters in ML cannot be respected and a series of complex mathematical operations in OR lead to the increased accumulative errors. Essentially, the improved precision on the prediction parameters may not have a positive correlation on the final solution due to the side-effect from the decoupled design. In this paper, we propose a novel approach for solving resource allocation problems to mitigate the side-effects. Our key intuition is that we introduce the decision factor to establish a bridge between ML and OR such that the solution can be directly obtained in OR by only performing the sorting or comparison operations on the decision factor. Furthermore, we design a customized loss function that can conduct direct heterogeneous causal learning on the decision factor, an unbiased estimation of which can be guaranteed when the loss converges. As a case study, we apply our approach to two crucial problems in marketing: the binary treatment assignment problem and the budget allocation problem with multiple treatments. Both large-scale simulations and online A/B Tests demonstrate that our approach achieves significant improvement compared with state-of-the-art.

Kaiwen Tang, Jiaqi Zheng, Yuze Jin et al.

Time-series forecasting in domains like traffic management and industrial monitoring often requires real-time, energy-efficient processing on edge devices with limited resources. Spiking neural networks (SNNs) offer event-driven computation and ultra-low power and have been proposed for use in this space. Unfortunately, existing SNN-based time-series forecasters often use complex transformer blocks. To address this issue, we propose SpikySpace, a spiking state-space model (SSM) that reduces the quadratic cost in the attention block to linear time via spiking selective scanning. Further, we introduce PTsoftplus and PTSiLU, two efficient approximations of SiLU and Softplus that replace costly exponential and division operations with simple bit-shifts. Evaluated on four multivariate time-series benchmarks, SpikySpace outperforms the leading SNN in terms of accuracy by up to 3.0% while reducing energy consumption by over 96.1%. As the first fully spiking state-space model, SpikySpace bridges neuromorphic efficiency with modern sequence modeling, opening a practical path toward efficient time series forecasting systems. Our code is available at https://anonymous.4open.science/r/SpikySpace.

Hongxuan Zhang, Zhining Liu, Yao Zhao et al.

In this work, we propose FastCoT, a model-agnostic framework based on parallel decoding without any further training of an auxiliary model or modification to the LLM itself. FastCoT uses a size-varying context window whose size changes with position to conduct parallel decoding and auto-regressive decoding simultaneously, thus fully utilizing GPU computation resources. In FastCoT, the parallel decoding part provides the LLM with a quick glance of the future composed of approximate tokens, which could lead to faster answers compared to regular autoregressive decoding used by causal transformers. We also provide an implementation of parallel decoding within LLM, which supports KV-cache generation and batch processing. Through extensive experiments, we demonstrate that FastCoT saves inference time by nearly 20% with only a negligible performance drop compared to the regular approach. Additionally, we show that the context window size exhibits considerable robustness for different tasks.

Peiran Xu, Jiaqi Zheng, Yadong Mu

This paper focuses on embodied task planning, where an agent acquires visual observations from the environment and executes atomic actions to accomplish a given task. Although recent Vision-Language Models (VLMs) have achieved impressive results in multimodal understanding and reasoning, their performance remains limited when applied to embodied planning that involves multi-turn interaction, long-horizon reasoning, and extended context analysis. To bridge this gap, we propose RoboAgent, a capability-driven planning pipeline in which the model actively invokes different sub-capabilities. Each capability maintains its own context, and produces intermediate reasoning results or interacts with the environment according to the query given by a scheduler. This framework decomposes complex planning into a sequence of basic vision-language problems that VLMs can better address, enabling a more transparent and controllable reasoning process. The scheduler and all capabilities are implemented with a single VLM, without relying on external tools. To train this VLM, we adopt a multi-stage paradigm that consists of: (1) behavior cloning with expert plans, (2) DAgger training using trajectories collected by the model, and (3) reinforcement learning guided by an expert policy. Across these stages, we exploit the internal information of the environment simulator to construct high-quality supervision for each capability, and we further introduce augmented and synthetic data to enhance the model's performance in more diverse scenarios. Extensive experiments on widely used embodied task planning benchmarks validate the effectiveness of the proposed approach. Our codes will be available at https://github.com/woyut/RoboAgent_CVPR26.

Mingliang Hou, Yinuo Wang, Teng Guo et al.

Cognitive diagnosis models (CDMs) are pivotal for creating fine-grained learner profiles in modern intelligent education platforms. However, these models are trained on sensitive student data, raising significant privacy concerns. While membership inference attacks (MIA) have been studied in various domains, their application to CDMs remains a critical research gap, leaving their privacy risks unquantified. This paper is the first to systematically investigate MIA against CDMs. We introduce a novel and realistic grey box threat model that exploits the explainability features of these platforms, where a model's internal knowledge state vectors are exposed to users through visualizations such as radar charts. We demonstrate that these vectors can be accurately reverse-engineered from such visualizations, creating a potent attack surface. Based on this threat model, we propose a profile-based MIA (P-MIA) framework that leverages both the model's final prediction probabilities and the exposed internal knowledge state vectors as features. Extensive experiments on three real-world datasets against mainstream CDMs show that our grey-box attack significantly outperforms standard black-box baselines. Furthermore, we showcase the utility of P-MIA as an auditing tool by successfully evaluating the efficacy of machine unlearning techniques and revealing their limitations.

Mingliang Hou, Yinuo Wang, Teng Guo et al.

The need to remove specific student data from cognitive diagnosis (CD) models has become a pressing requirement, driven by users' growing assertion of their "right to be forgotten". However, existing CD models are largely designed without privacy considerations and lack effective data unlearning mechanisms. Directly applying general purpose unlearning algorithms is suboptimal, as they struggle to balance unlearning completeness, model utility, and efficiency when confronted with the unique heterogeneous structure of CD models. To address this, our paper presents the first systematic study of the data unlearning problem for CD models, proposing a novel and efficient algorithm: hierarchical importanceguided forgetting (HIF). Our key insight is that parameter importance in CD models exhibits distinct layer wise characteristics. HIF leverages this via an innovative smoothing mechanism that combines individual and layer, level importance, enabling a more precise distinction of parameters associated with the data to be unlearned. Experiments on three real world datasets show that HIF significantly outperforms baselines on key metrics, offering the first effective solution for CD models to respond to user data removal requests and for deploying high-performance, privacy preserving AI systems

Hao Zhou, Rongxiao Huang, Shaoming Li et al.

Marketing optimization plays an important role to enhance user engagement in online Internet platforms. Existing studies usually formulate this problem as a budget allocation problem and solve it by utilizing two fully decoupled stages, i.e., machine learning (ML) and operation research (OR). However, the learning objective in ML does not take account of the downstream optimization task in OR, which causes that the prediction accuracy in ML may be not positively related to the decision quality. Decision Focused Learning (DFL) integrates ML and OR into an end-to-end framework, which takes the objective of the downstream task as the decision loss function and guarantees the consistency of the optimization direction between ML and OR. However, deploying DFL in marketing is non-trivial due to multiple technological challenges. Firstly, the budget allocation problem in marketing is a 0-1 integer stochastic programming problem and the budget is uncertain and fluctuates a lot in real-world settings, which is beyond the general problem background in DFL. Secondly, the counterfactual in marketing causes that the decision loss cannot be directly computed and the optimal solution can never be obtained, both of which disable the common gradient-estimation approaches in DFL. Thirdly, the OR solver is called frequently to compute the decision loss during model training in DFL, which produces huge computational cost and cannot support large-scale training data. In this paper, we propose a decision focused causal learning framework (DFCL) for direct counterfactual marketing optimization, which overcomes the above technological challenges. Both offline experiments and online A/B testing demonstrate the effectiveness of DFCL over the state-of-the-art methods. Currently, DFCL has been deployed in several marketing scenarios in Meituan, one of the largest online food delivery platform in the world.

Minyuan Zhou, Yuning Chen, Jiaqi Zheng et al.

Operating large-scale anycast networks is challenging because client-to-site mappings often misalign with operator's expectation due to opaque inter-domain routing. We present AnyPro, the first system to unlock the full potential of AS-path prepending (ASPP), efficiently deriving globally optimal configurations to steer clients toward performance-optimal sites at scale. AnyPro first employs an efficient polling mechanism to identify all clients sensitive to ASPP. By analyzing the routing changes during the process, the system derives a set of ASPP constraints that guide client traffic toward the desired sites. We then formulate the anycast optimization problem as a constraint-based program and compute optimal ASPP configurations. Extensive evaluation on a global testbed with 20 PoPs demonstrates the effectiveness of AnyPro: it reduces the 90th percentile latency by 37.7% compared to baseline configurations without ASPP. Furthermore, we show that AnyPro can be integrated with PoP-level anycast optimization techniques to achieve additional performance gains.

Jiaqi Zheng, Qing Ling, Yerong Feng

Although deep learning models have demonstrated remarkable potential in weather prediction, most of them overlook either the \textbf{physics} of the underlying weather evolution or the \textbf{topology} of the Earth's surface. In light of these disadvantages, we develop PASSAT, a novel Physics-ASSisted And Topology-informed deep learning model for weather prediction. PASSAT attributes the weather evolution to two key factors: (i) the advection process that can be characterized by the advection equation and the Navier-Stokes equation; (ii) the Earth-atmosphere interaction that is difficult to both model and calculate. PASSAT also takes the topology of the Earth's surface into consideration, other than simply treating it as a plane. With these considerations, PASSAT numerically solves the advection equation and the Navier-Stokes equation on the spherical manifold, utilizes a spherical graph neural network to capture the Earth-atmosphere interaction, and generates the initial velocity fields that are critical to solving the advection equation from the same spherical graph neural network. In the $5.625^\circ$-resolution ERA5 data set, PASSAT outperforms both the state-of-the-art deep learning-based weather prediction models and the operational numerical weather prediction model IFS T42. Code and checkpoint are available at https://github.com/Yumenomae/PASSAT_5p625.

Zixuan Yang, Jiaqi Zheng, Guihai Chen

Ensuring verifiable and interpretable safety of deep reinforcement learning (DRL) is crucial for its deployment in real-world applications. Existing approaches like verification-in-the-loop training, however, face challenges such as difficulty in deployment, inefficient training, lack of interpretability, and suboptimal performance in property satisfaction and reward performance. In this work, we propose a novel verification-driven interpretation-in-the-loop framework Reintrainer to develop trustworthy DRL models, which are guaranteed to meet the expected constraint properties. Specifically, in each iteration, this framework measures the gap between the on-training model and predefined properties using formal verification, interprets the contribution of each input feature to the model's output, and then generates the training strategy derived from the on-the-fly measure results, until all predefined properties are proven. Additionally, the low reusability of existing verifiers and interpreters motivates us to develop Reinfier, a general and fundamental tool within Reintrainer for DRL verification and interpretation. Reinfier features breakpoints searching and verification-driven interpretation, associated with a concise constraint-encoding language DRLP. Evaluations demonstrate that Reintrainer outperforms the state-of-the-art on six public benchmarks in both performance and property guarantees. Our framework can be accessed at https://github.com/Kurayuri/Reinfier.

Hao Zhou, Kun Sun, Shaoming Li et al.

Online controlled experiments play a crucial role in enabling data-driven decisions across a wide range of companies. Variance reduction is an effective technique to improve the sensitivity of experiments, achieving higher statistical power while using fewer samples and shorter experimental periods. However, typical variance reduction methods (e.g., regression-adjusted estimators) are built upon the intuitional assumption of Gaussian distributions and cannot properly characterize the real business metrics with heavy-tailed distributions. Furthermore, outliers diminish the correlation between pre-experiment covariates and outcome metrics, greatly limiting the effectiveness of variance reduction. In this paper, we develop a novel framework that integrates the Student's t-distribution with machine learning tools to fit heavy-tailed metrics and construct a robust average treatment effect estimator in online controlled experiments, which we call STATE. By adopting a variational EM method to optimize the loglikehood function, we can infer a robust solution that greatly eliminates the negative impact of outliers and achieves significant variance reduction. Moreover, we extend the STATE method from count metrics to ratio metrics by utilizing linear transformation that preserves unbiased estimation, whose variance reduction is more complex but less investigated in existing works. Finally, both simulations on synthetic data and long-term empirical results on Meituan experiment platform demonstrate the effectiveness of our method. Compared with the state-of-the-art estimators (CUPAC/MLRATE), STATE achieves over 50% variance reduction, indicating it can reach the same statistical power with only half of the observations, or half the experimental duration.

Kaiwen Tang, Di Yu, Jiaqi Zheng et al.

Spiking neural networks (SNNs) are promising for edge sensing due to their event-driven computation and temporal filtering capability. However, standard leaky integrate-and-fire (LIF) neurons communicate only through binary spikes, which severely limit representational capacity. Existing multi-level spiking neurons improve information transmission, but often rely on uniform quantization that mismatches membrane-potential distributions or introduces costly synaptic multiplications. In this paper, we propose ShiftLIF, a multi-level spiking neuron that maps membrane potentials to a logarithmically spaced power-of-two spike set. This design provides finer representation in the small-amplitude regime, where membrane potentials are densely concentrated, while enabling multiplier-free synaptic computation through bit-shift and accumulation operations. As a result, ShiftLIF improves spike-level expressiveness without sacrificing the hardware-friendly nature of standard SNN computation. We evaluate ShiftLIF on 10 datasets spanning wireless, acoustic, motion, and visual sensing tasks. Results show that ShiftLIF consistently matches or exceeds the accuracy of existing multi-level spiking neurons while maintaining synaptic energy consumption close to standard binary LIF. These results indicate that ShiftLIF provides a favorable accuracy-efficiency trade-off for cross-modal edge sensing.

Shiju Zhao, Junhao Hu, Rongxiao Huang et al.

The context caching technique is employed to accelerate the Multimodal Large Language Model (MLLM) inference by prevailing serving platforms currently. However, this approach merely reuses the Key-Value (KV) cache of the initial sequence of prompt, resulting in full KV cache recomputation even if the prefix differs slightly. This becomes particularly inefficient in the context of interleaved text and images, as well as multimodal retrieval-augmented generation. This paper proposes position-independent caching as a more effective approach for multimodal information management. We have designed and implemented a caching system, named MPIC, to address both system-level and algorithm-level challenges. MPIC stores the KV cache on local disks when receiving multimodal data, and calculates and loads the KV cache in parallel during inference. To mitigate accuracy degradation, we have incorporated the integrated reuse and recompute mechanism within the system. The experimental results demonstrate that MPIC can achieve up to 54\% reduction in response time and 2$\times$ improvement in throughput compared to existing context caching systems, while maintaining negligible or no accuracy loss.

Hongxuan Zhang, Yao Zhao, Jiaqi Zheng et al.

The emergence of long-context text applications utilizing large language models (LLMs) has presented significant scalability challenges, particularly in memory footprint. The linear growth of the Key-Value (KV) cache responsible for storing attention keys and values to minimize redundant computations can lead to substantial increases in memory consumption, potentially causing models to fail to serve with limited memory resources. To address this issue, we propose a novel approach called Cache Sparse Representation (CSR), which converts the KV cache by transforming the dense Key-Value cache tensor into sparse indexes and weights, offering a more memory-efficient representation during LLM inference. Furthermore, we introduce NeuralDict, a novel neural network-based method for automatically generating the dictionary used in our sparse representation. Our extensive experiments demonstrate that CSR achieves performance comparable to state-of-the-art KV cache quantization algorithms while maintaining robust functionality in memory-constrained environments.

Shuli Zhang, Hao Zhou, Jiaqi Zheng et al.

Online Internet platforms require sophisticated marketing strategies to optimize user retention and platform revenue -- a classical resource allocation problem. Traditional solutions adopt a two-stage pipeline: machine learning (ML) for predicting individual treatment effects to marketing actions, followed by operations research (OR) optimization for decision-making. This paradigm presents two fundamental technical challenges. First, the prediction-decision misalignment: Conventional ML methods focus solely on prediction accuracy without considering downstream optimization objectives, leading to improved predictive metrics that fail to translate to better decisions. Second, the bias-variance dilemma: Observational data suffers from multiple biases (e.g., selection bias, position bias), while experimental data (e.g., randomized controlled trials), though unbiased, is typically scarce and costly -- resulting in high-variance estimates. We propose Bi-level Decision-Focused Causal Learning (Bi-DFCL) that systematically addresses these challenges. First, we develop an unbiased estimator of OR decision quality using experimental data, which guides ML model training through surrogate loss functions that bridge discrete optimization gradients. Second, we establish a bi-level optimization framework that jointly leverages observational and experimental data, solved via implicit differentiation. This novel formulation enables our unbiased OR estimator to correct learning directions from biased observational data, achieving optimal bias-variance tradeoff. Extensive evaluations on public benchmarks, industrial marketing datasets, and large-scale online A/B tests demonstrate the effectiveness of Bi-DFCL, showing statistically significant improvements over state-of-the-art. Currently, Bi-DFCL has been deployed at Meituan, one of the largest online food delivery platforms in the world.

Dali Zhu, Wenli Zhang, Hualin Zeng et al.

Remote photoplethysmography (rPPG) technique extracts blood volume pulse (BVP) signals from subtle pixel changes in video frames. This study introduces rFaceNet, an advanced rPPG method that enhances the extraction of facial BVP signals with a focus on facial contours. rFaceNet integrates identity-specific facial contour information and eliminates redundant data. It efficiently extracts facial contours from temporally normalized frame inputs through a Temporal Compressor Unit (TCU) and steers the model focus to relevant facial regions by using the Cross-Task Feature Combiner (CTFC). Through elaborate training, the quality and interpretability of facial physiological signals extracted by rFaceNet are greatly improved compared to previous methods. Moreover, our novel approach demonstrates superior performance than SOTA methods in various heart rate estimation benchmarks.