arXiv subjects

Ying Sun

Publications and source records attributed to Ying Sun.

At least 37 records · Page 2Linked to original sources

arXiv2026

Escaping the KL Agreement Trap in On-Policy Distillation

On-policy distillation (OPD) provides dense token-level supervision by asking a teacher to score student-generated rollouts. However, when the student drifts into an unrecoverable prefix, the teacher may locally agree with the degraded state, producing low reverse KL but little corrective training signal. We identify this persistent regime as a low-KL agreement trap. Further analyses show that tokens during and after such traps produce less useful supervision signals. We propose KAT (KL Agreement Trap Termination), an online OPD termination rule that detects persistent low-KL agreement with a dynamic training-adaptive threshold. By filtering weak supervision from degenerate agreement, KAT improves avg@k accuracy by 2.66% and pass@k by 3.43% across four mathematical benchmarks, while reducing average rollout length by 59.73%.

cs.LG↗

arXiv2026

Chameleon: Control-Indexed Prospective Memory for Visuomotor Manipulation

Robots often observe information that determines a future action long before that action is executed. In a shell game, for example, a robot first sees which cup hides the ball, watches the cups move, and only later needs to choose the correct cup. The final observation alone is not enough for a decision: the correct action depends on an earlier event. We refer to this temporal gap as observation-action delay. It makes memory a policy-facing problem: a policy must keep similar histories distinct, retrieve the past event relevant to the current decision, and convert that recall into an action-ready state. We call these requirements separability, addressability, and prospectiveness. We introduce Chameleon, a ~60M visuomotor policy for control-indexed prospective memory. Chameleon writes embodied event memory, preserves separable histories, retrieves control-relevant traces, and trains the resulting working state to be prospective. We also introduce Camo-Dataset, a real-robot benchmark that isolates observation-action delay by making the decision scene visually ambiguous, so the correct action must be inferred from earlier observations. Chameleon improves decision/end-to-end success on Camo-Dataset from 22.5%/21.3% to 80.8%/71.3%. On public long-horizon memory benchmarks, it achieves 87.1% +/- 0.8% on LIBERO-10, 97.3% +/- 4.5% on MemoryBench, and 75.1% +/- 1.4% on MIKASA-Robo, setting the state of the art for same-size models and exceeding multiple larger VLA baselines under the reported protocols. Probes and ablations show that Chameleon learns separable, addressable, and prospective memory, and that these properties drive its performance gains.

cs.RO↗

arXiv2026

Mid-infrared photon counting and resolving via efficient frequency upconversion

Optical detectors with single-photon sensitivity and large dynamic range would facilitate a variety of applications. Especially, the capability of extending operation wavelengths into the mid-infrared region is highly attractive. Here we implement a mid-infrared frequency upconversion detector for counting and resolving photons at 3 $μ$m. Thanks to the spectro-temporal engineering of the involved optical fields, the mid-infrared photons could be spectrally translated into the visible band with a conversion efficiency of 80\%. In combination with a silicon avalanche photodiode, we obtained unprecedented performances with a high overall detection efficiency of 37\% and a low noise equivalent power of 1.8$\times$10$^{-17}$ W/Hz$^{1/2}$. Furthermore, photon-number-resolving detection at mid-infrared wavelengths was demonstrated, for the first time to our knowledge, with a multi-pixel photon counter. The implemented upconversion detector exhibited a maximal resolving photon number up to 9 with a noise probability per pulse of 0.14\% at the peak detection efficiency. The achieved photon counting and resolving performance might open up new possibilities in trace molecule spectroscopy, sensitive biochemical sensing, and free-space communications, among others.

physics.optics↗

arXiv2026

IRDS: Interpretable RLVR Data Selection via Verifier-Coupled Sparse Autoencoder Coverage

Reinforcement learning with verifiable rewards (RLVR) has become a key technique for en- hancing LLM reasoning, yet its data ineffi- ciency remains a major bottleneck. Existing methods address this problem only partially, each missing at least one of subset-level cov- erage, verifier signal use, or interpretability. To address this gap, we present IRDS (Inter- pretable RLVR Data Selection), which selects RLVR training instances on a sparse autoen- coder (SAE) cluster basis so the selection itself is auditable on recognizable problem motifs. To select instances the model both fails on and can still learn from, we introduce a verifier- coupled coverage objective on the SAE basis and solve it by greedy log-determinant max- imization. Experiments on three instruction- tuned models and six math reasoning bench- marks show that IRDS achieves the highest overall accuracy, exceeding the strongest base- line by +3.9/+4.0 pp on the two Qwen models and by +0.5 pp on Llama-3.1-8B, while run- ning an order of magnitude cheaper than the trajectory-based baseline.

cs.LG↗

arXiv2026

SAE-FD: Sparse Autoencoder Feature Distillation for Continual Learning of Large Language Models

Continual learning enables large language models to adapt to evolving tasks without retraining from scratch, yet catastrophic forgetting remains a central obstacle. Among continual learning methods, regularization-based approaches are widely used to constrain model updates and reduce forgetting, operating in weight space, gradient space, or output space. However, these dense representation spaces suffer from feature superposition, where multiple concepts are encoded in overlapping dimensions, making it difficult to selectively protect previously learned knowledge without impeding new-task learning. To address this issue, we propose \method (Sparse Autoencoder Feature Distillation), which anchors model representations in the sparse feature space of a pre-trained Sparse Autoencoder, where dense activations are decomposed into a sparse overcomplete basis that reduces representational entanglement, enabling more targeted regularization with less interference to new-task learning. Experiments on two continual learning benchmarks across three model architectures show that \method consistently outperforms existing regularization-based methods, achieving up to 52.70% average accuracy with only -0.46 backward transfer.

cs.LG↗

arXiv2026

HDMoE: A Hierarchical Decoupling-Fusion Mixture-of-Experts Framework for Multimodal Cancer Survival Prediction

Multimodal survival prediction, a crucial yet challenging task, demands the integration of multimodal medical data (\eg Whole Slide Images (WSIs) and Genomic Profiles) to achieve accurate prognostic modeling. Given the inherent heterogeneity across modalities, the feature decoupling-fusion paradigm has emerged as a dominant approach. However, these methods have the following shortcomings: (1) fail to reduce the redundant information of modality features before decoupling, which negatively affects the feature decoupling and fusion effect;(2) lack the ability to model the fine-grained relationships of the features and capture the local information interactions between intra- and inter-modality features. To address these issues, we propose a \underline{H}ierarchical \underline{D}ecoupling-Fusion \underline{M}ixture-\underline{o}f-\underline{E}xperts (HDMoE) framework with two levels of MoE and \underline{R}andom \underline{F}eature \underline{R}eorganization (RFR) modules.In the first-level MoE, shared experts and routed experts are employed to remove redundant information and extract fine-grained specific features within each modality, while the second-level MoE facilitates fine-grained inter-modality feature decoupling. Besides, we design two RFR modules following each level of MoE to finely fuse intra- and inter-modality features, which can help the model capture more fine-grained relationships between modalities. Extensive experimental results on our private Liver Cancer (LC) and three TCGA public datasets confirm the effectiveness of our proposed method. Codes are available at https://github.com/ZJUMAI/HDMoE.

cs.CV↗

arXiv2026

Amortized Neural Clustering of Time Series based on Statistical Features

This paper introduces an algorithm-agnostic approach to feature-based time series clustering via amortized neural inference. By training neural networks to approximate the optimal partitioning rule from simulated data, the proposed framework reduces reliance on conventional clustering methods, such as $K$-means, $K$-medoids, or hierarchical clustering, and their associated objective functions and heuristics. Leveraging statistical features, such as autocorrelations and quantile autocorrelations, the approach learns a data-driven affinity structure from which clustering partitions can be recovered, without requiring explicit prior specification of cluster shapes or structures. In addition, one version of the method can automatically determine the number of clusters, avoiding ad-hoc selection procedures. Comprehensive empirical studies show that the proposed framework achieves competitive or superior clustering accuracy relative to traditional methods, even in challenging scenarios where competing techniques are provided with the true number of clusters. An application to financial time series of stock returns illustrates its practical utility. By reducing the need for algorithm selection and calibration, the proposed framework opens new possibilities for automated, adaptive, and data-driven clustering of temporal data across scientific and industrial domains.

stat.ML↗

arXiv2026

SLIM: Sparse Latent Steering for Interpretable and Property-Directed LLM-Based Molecular Editing

Large language models possess strong chemical reasoning capabilities, making them effective molecular editors. However, property-relevant information is implicitly entangled across their dense hidden states, providing no explicit handle for property control: a substantial fraction of edits fail to improve or even degrade target properties. To address these issues, we propose SLIM (Sparse Latent Interpretable Molecular editing), a plug-and-play framework that decomposes the editor's hidden states into sparse, property-aligned features via a Sparse Autoencoder with learnable importance gates. Steering in this sparse feature space precisely activates property-relevant dimensions, improving editing success rate without modifying model parameters. The same sparse basis further supports interpretable analysis of editing behavior. Experiments on the MolEditRL benchmark across four model architectures and eight molecular properties show consistent gains over baselines, with improvements of up to 42.4 points.

cs.LG↗

arXiv2026

White noise testing for functional time series via functional quantile autocorrelation

We introduce a novel class of nonlinear tests for serial dependence in functional time series, grounded in the functional quantile autocorrelation framework. Unlike traditional approaches based on the classical autocovariance kernel, the functional quantile autocorrelation framework leverages quantile-based excursion sets to robustly capture temporal dependence within infinite-dimensional functional data, accommodating potential outliers and complex nonlinear dependencies. We propose omnibus test statistics and study their asymptotic properties under both known and estimated quantile curves, establishing their asymptotic distribution and consistency under mild assumptions. In particular, no moment conditions are required for the validity of the tests. Extensive simulations and an application to high-frequency financial functional time series demonstrate the methodology's effectiveness, reliably detecting complex serial dependence with superior power relative to several existing tests. This work expands the toolkit for functional time series, providing a robust framework for inference in settings where traditional methods may fail.

stat.ME↗

arXiv2026

Layer-Guided UAV Tracking: Enhancing Efficiency and Occlusion Robustness

Visual object tracking (VOT) plays a pivotal role in unmanned aerial vehicle (UAV) applications. Addressing the trade-off between accuracy and efficiency, especially under challenging conditions like unpredictable occlusion, remains a significant challenge. This paper introduces LGTrack, a unified UAV tracking framework that integrates dynamic layer selection, efficient feature enhancement, and robust representation learning for occlusions. By employing a novel lightweight Global-Grouped Coordinate Attention (GGCA) module, LGTrack captures long-range dependencies and global contexts, enhancing feature discriminability with minimal computational overhead. Additionally, a lightweight Similarity-Guided Layer Adaptation (SGLA) module replaces knowledge distillation, achieving an optimal balance between tracking precision and inference efficiency. Experiments on three datasets demonstrate LGTrack's state-of-the-art real-time speed (258.7 FPS on UAVDT) while maintaining competitive tracking accuracy (82.8\% precision). Code is available at https://github.com/XiaoMoc/LGTrack

cs.CV↗

arXiv2026

Discrete Preference Learning for Personalized Multimodal Generation

The emergence of generative models enables the creation of texts and images tailored to users' preferences. Existing personalized generative models have two critical limitations: lacking a dedicated paradigm for accurate preference modeling, and generating unimodal content despite real-world multimodal-driven user interactions. Therefore, we propose personalized multimodal generation, which captures modal-specific preferences via a dedicated preference model from multimodal interactions, and then feeds them into downstream generators for personalized multimodal content. However, this task presents two challenges: (1) Gap between continuous preferences from dedicated modeling and discrete token inputs intrinsic to generator architectures; (2) Potential inconsistency between generated images and texts. To tackle these, we present a two-stage framework called Discrete Preference learning for Personalized Multimodal Generation (DPPMG). In the first stage, to accurately learn discrete modal-specific preferences, we introduce a modal-specific graph neural network (a dedicated preference model) to learn users' modal-specific preferences, which preferences are then quantized into discrete preference tokens. In the second stage, the discrete modal-specific preference tokens are injected into downstream text and image generators. To further enhance cross-modal consistency while preserving personalization, we design a cross-modal consistent and personalized reward to fine-tune token-associated parameters. Extensive experiments on two real-world datasets demonstrate the effectiveness of our model in generating personalized and consistent multimodal content.

cs.IR↗

arXiv2026

Latent-Condensed Transformer for Efficient Long Context Modeling

Large language models (LLMs) face significant challenges in processing long contexts due to the linear growth of the key-value (KV) cache and quadratic complexity of self-attention. Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. However, sparse methods cannot operate natively on MLA's compressed latent structure, missing opportunities for joint optimization. In this paper, we propose Latent-Condensed Attention (LCA), which directly condenses context within MLA's latent space, where the representation is disentangled into semantic latent vectors and positional keys. LCA separately aggregates semantic vectors via query-aware pooling and preserves positional keys via anchor selection. This approach jointly reduces both computational cost and KV cache without adding parameters. Beyond MLA, LCA's design is architecture-agnostic and readily extends to other attention mechanisms such as GQA. Theoretically, we prove a length-independent error bound. Experiments show LCA achieves up to 2.5$\times$ prefilling speedup and 90% KV cache reduction at 128K context while maintaining competitive performance.

cs.CL↗

arXiv2026

Forecasting Multivariate Time Series under Predictive Heterogeneity: A Validation-Driven Clustering Framework

We study adaptive pooling under predictive heterogeneity in high-dimensional multivariate time series forecasting, where global models improve statistical efficiency but may fail to capture heterogeneous predictive structure, while naive specialization can induce negative transfer. We formulate adaptive pooling as a statistical decision problem and propose a validation-driven framework that determines when and how specialization should be applied. Rather than grouping series based on representation similarity, we define partitions through out-of-sample predictive performance, thereby aligning data organization with predictive risk, defined as expected out-of-sample loss and approximated via validation error. Cluster assignments are iteratively updated using validation losses for both point (Huber) and probabilistic (pinball) forecasting, improving robustness to heavy-tailed errors and local anomalies. To ensure reliability, we introduce a leakage-free fallback mechanism that reverts to a global model whenever specialization fails to improve validation performance, providing a safeguard against performance degradation under a strict training-validation-test protocol. Experiments on large-scale traffic datasets demonstrate consistent improvements over strong baselines while avoiding degradation when heterogeneity is weak. Overall, the proposed framework provides a principled and practically reliable approach to adaptive pooling in high-dimensional forecasting problems.

stat.ME↗

arXiv2026

Efficient Handwriting-Based Alzheimer,s Disease Diagnosis Using a Low-Rank Mixture of Experts Deep Learning Framework

Early and reliable detection of Alzheimer's disease (AD) is crucial for timely clinical intervention and improved patient management. It also supports the evaluation of emerging therapeutic strategies. In this paper, we propose a Low-Rank Mixture of Experts (LoRA-MoE) deep learning framework for Alzheimer's disease diagnosis based on handwriting analysis. Handwriting signals provide a non-invasive and scalable digital biomarker that captures subtle cognitive-motor impairments associated with early AD progression. The proposed architecture allows multiple experts to specialize in different handwriting patterns while sharing a common base network. This design enables efficient learning of general representations while reducing interference between experts. Each expert is equipped with lightweight low-rank adapters. This mechanism significantly reduces the number of trainable parameters compared with standard Mixture of Experts (MoE) models and improves training stability. The proposed framework is evaluated on the Diagnosis AlzheimeR WIth haNdwriting (DARWIN) dataset. Extensive experiments are conducted, including ablation studies on key architectural parameters such as hidden dimension size, number of experts, and LoRA rank. The method is compared with multilayer perceptron (MLP) and conventional MoE architectures. In addition, stacking ensemble strategies (StackMean and StackMax) are investigated to improve robustness and predictive performance. Experimental results show that the LoRA-MoE framework achieves powerful diagnostic performance while activating significantly fewer parameters during inference. These results highlight the potential of the proposed approach as an accurate and computationally efficient solution for handwriting-based Alzheimer's disease screening and digital health applications.

cs.LG↗

arXiv2026

Biogeochemistry-Informed Neural Network (BINN) for Improving Accuracy of Model Prediction and Scientific Understanding of Soil Organic Carbon

The increasing availability of large-scale observational data and the rapid development of artificial intelligence (AI) provide unprecedented opportunities to enhance our understanding of the global carbon cycle and other biogeochemical processes. However, retrieving mechanistic knowledge from these large-scale data remains a challenge. Here, we develop a Biogeochemistry-Informed Neural Network (BINN) that seamlessly integrates a vectorized process-based soil carbon cycle model (i.e., Community Land Model version 5, CLM5) into a neural network (NN) structure to examine mechanisms governing soil organic carbon (SOC) storage from big data. BINN demonstrates high accuracy in retrieving biogeochemical parameter values from synthetic data in a parameter recovery experiment. Furthermore, by incorporating Monte Carlo (MC) dropout to generate posterior distributions, we demonstrate that BINN can effectively quantify uncertainty in estimated parameters. We use BINN to predict six major processes (or components in process-based models) regulating the soil carbon cycle from 25,925 observed SOC profiles across the contiguous US and compare them with the same processes previously retrieved by a Bayesian inference-based PROcess-guided deep learning and DAta-driven modeling (PRODA) approach. The good agreement between the spatial patterns retrieved by BINN and PRODA (average correlation coefficient = 0.86) suggests that BINN's ability of capturing mechanistic knowledge is consistent with the established Bayesian-based methods. Additionally, the integration of neural networks and process-based models in BINN improves computational efficiency by more than 50 times over PRODA. We conclude that BINN is an efficient framework that harnesses the power of both AI, large-scale data, and process-based modeling to understand large scale soil carbon cycle.

physics.geo-ph↗

arXiv2026

Scaled Block Vecchia Approximation for High-Dimensional Gaussian Process Emulation on GPUs

Emulating computationally intensive scientific simulations is crucial for enabling uncertainty quantification, optimization, and informed decision-making at scale. Gaussian Processes (GPs) offer a flexible and data-efficient foundation for statistical emulation, but their poor scalability limits applicability to large datasets. We introduce the Scaled Block Vecchia (SBV) algorithm for distributed GPU-based systems. SBV integrates the Scaled Vecchia approach for anisotropic input scaling with the Block Vecchia (BV) method to reduce computational and memory complexity while leveraging GPU acceleration techniques for efficient linear algebra operations. To the best of our knowledge, this is the first distributed implementation of any Vecchia-based GP variant. Our implementation employs MPI for inter-node parallelism and the MAGMA library for GPU-accelerated batched matrix computations. We demonstrate the scalability and efficiency of the proposed algorithm through experiments on synthetic and real-world workloads, including a 50M point simulation from a respiratory disease model. SBV achieves near-linear scalability on up to 512 A100 and GH200 GPUs, handles 2.56B points, and reduces energy use relative to exact GP solvers, establishing SBV as a scalable and energy-efficient framework for emulating large-scale scientific models on GPU-based distributed systems.

cs.DC↗

arXiv2026

Discovering Decoupled Functional Modules in Large Language Models

Understanding the internal functional organization of Large Language Models (LLMs) is crucial for improving their trustworthiness and performance. However, how LLMs organize different functions into modules remains highly unexplored. To bridge this gap, we formulate a functional module discovery problem and propose an Unsupervised LLM Cross-layer MOdule Discovery (ULCMOD) framework that simultaneously disentangles the large set of neurons in the entire LLM into modules while discovering the topics of input samples related to these modules. Our framework introduces a novel objective function and an efficient Iterative Decoupling (IterD) algorithm. Extensive experiments show that our method discovers high-quality, disentangled modules that capture more meaningful semantic information and achieve superior performance in various downstream tasks. Moreover, our qualitative analysis reveals that the discovered modules show semantic coherence, correspond to interpretable specializations, and a clear spatial and hierarchical organization within the LLM. Our work provides a novel tool for interpreting the functional modules of LLMs, filling a critical blank in LLM's interpretability research.

cs.LG↗

arXiv2026

GoldenStart: Q-Guided Priors and Entropy Control for Distilling Flow Policies

Flow-matching policies hold great promise for reinforcement learning (RL) by capturing complex, multi-modal action distributions. However, their practical application is often hindered by prohibitive inference latency and ineffective online exploration. Although recent works have employed one-step distillation for fast inference, the structure of the initial noise distribution remains an overlooked factor that presents significant untapped potential. This overlooked factor, along with the challenge of controlling policy stochasticity, constitutes two critical areas for advancing distilled flow-matching policies. To overcome these limitations, we propose GoldenStart (GSFlow), a policy distillation method with Q-guided priors and explicit entropy control. Instead of initializing generation from uninformed noise, we introduce a Q-guided prior modeled by a conditional VAE. This state-conditioned prior repositions the starting points of the one-step generation process into high-Q regions, effectively providing a "golden start" that shortcuts the policy to promising actions. Furthermore, for effective online exploration, we enable our distilled actor to output a stochastic distribution instead of a deterministic point. This is governed by entropy regularization, allowing the policy to shift from pure exploitation to principled exploration. Our integrated framework demonstrates that by designing the generative startpoint and explicitly controlling policy entropy, it is possible to achieve efficient and exploratory policies, bridging the generative models and the practical actor-critic methods. We conduct extensive experiments on offline and online continuous control benchmarks, where our method significantly outperforms prior state-of-the-art approaches. Code will be available at https://github.com/ZhHe11/GSFlow-RL.

cs.LG↗