arXiv subjects

Nan Qiao

Publications and source records attributed to Nan Qiao.

At least 19 recordsLinked to original sources

arXiv2026

COINS: Any-Stage-Valid and Utility-Oriented Sequential Conformal Prediction

Many predictive workflows update uncertainty as information is acquired and use intermediate reports to determine whether to stop or deploy further resources. We study conformal inference in this setting, treating the resulting prediction sequence as the inferential object. We require any-stage validity, which protects against miscoverage at any inspected stage, and use process-level utility to evaluate how efficiently the sequence supports downstream action. We propose a universal structural theory for constructing any-stage valid prediction sequences. Guided by it, we develop COINS, which coordinates calibration across stages by investing a common finite-sample rejection-count budget only among surviving augmented observations. Under exchangeability, COINS achieves finite-sample any-stage validity and produces prediction sets no larger than their matched Bonferroni counterparts at every stage. We further develop Vopt-COINS, which learns the stagewise allocation for a specified process-level utility, together with branchwise and localized extensions for heterogeneous acquisition pathways and test units. Simulations and a dermatological-diagnosis application confirm any-stage validity and demonstrate gains over Bonferroni and fixed allocations. The proposed methods also perform favorably in ordered score aggregation, viewed as a terminal-utility special case.

stat.ME↗

arXiv2026

Beyond Correctness: Benchmarking and Aligning Response Behaviors in Hybrid-Thinking MLLMs

Hybrid-thinking multimodal large language models (MLLMs) allow a single model to alternate between deliberative thinking and latency-efficient non-thinking inference. Although these modes differ in reasoning budget, their delivered responses should satisfy the same user-facing standard. Correctness alone may not characterize this response quality; we therefore evaluate task accuracy and response-pattern failures as complementary outcomes. We study this gap through \textbf{response-pattern alignment}: whether thinking and non-thinking interfaces preserve acceptable final-response behavior. We introduce \textbf{PatternEval}, a failure-enriched diagnostic benchmark comprising 2,415 multimodal prompts spanning visual perception and grounding, structured image understanding, and multimodal knowledge reasoning. PatternEval tests four recurrent failures: chain-of-thought leakage, response repetition, logical contradiction, and performative reasoning. Response-pattern failures are widespread across models from different providers, with non-thinking inference exhibiting substantially higher failure rates and thereby creating systematic misalignment between thinking and non-thinking interfaces. Motivated by this diagnosis, we develop \textbf{PatternRM}, a response-level reward model, and \textbf{PatternRL}, which introduces pattern-specific penalties during reinforcement learning. Experiments on Qwen3-VL-4B and Qwen3-VL-8B show that incorporating pattern-specific penalties into reinforcement learning can mitigate cross-mode misalignment while incurring a marginal task performance trade-off. Together, PatternEval and PatternRL provide an evaluation-and-training framework for aligning user-visible response patterns across hybrid-thinking interfaces.

cs.CV↗

arXiv2026

Enhancing Part-Level Point Grounding for Any Open-Source MLLMs

Visual grounding aims to associate free-form textual queries with specific regions in an image. While recent Multimodal Large Language Models (MLLMs) have demonstrated promising capabilities in this domain, they primarily excel at object-level grounding and often struggle with part-level grounding-an essential requirement for fine-grained tasks such as robotic manipulation. In this work, we introduce a general approach that equips any open-source MLLMs with accurate 2D part-level point grounding, offering a more direct alternative to conventional grounding representations. Our method leverages the attention mechanisms inherently present in MLLMs. By synthesizing text-conditioned, grounding-aware queries within intermediate layers via the proposed Q-Synth Module, we capture target-relevant attention patterns and refine them with a lightweight Attention-to-Point Decoder, which converts these patterns into a point-centric heatmap for final prediction. Notably, all original MLLM parameters are frozen, ensuring full preservation of their pre-trained capabilities. Experiments show that our design consistently improves part-level grounding accuracy across datasets and can be seamlessly integrated into any open-source MLLMs.

cs.CV↗

arXiv2026

AdamO: A Collapse-Suppressed Optimizer for Offline RL

Offline reinforcement learning (RL) can fail spectacularly when bootstrapped temporal-difference (TD) updates amplify their own errors, driving the critic toward extreme and unusable Q-values. A key counterintuitive insight of this work is that collapse is not only a property of the backup rule or network architecture: optimizer dynamics themselves can directly trigger or suppress instability. From a control-theoretic viewpoint, we model offline TD learning as a feedback system and analyze Adam-based critic updates. This yields a necessary and sufficient condition for stability of the induced local update dynamics: within the regime we analyze, these dynamics are stable if and only if the spectral radius of the corresponding update operator is strictly below one. Further analysis suggests that standard Adam updates can inadvertently distort the parameter geometry, motivating explicit orthogonality constraints to prevent TD error amplification. To this end, we propose AdamO, an Adam-based optimizer with a decoupled orthogonality correction regulated by a strict task-alignment budget. We prove that this design theoretically guarantees worst-case task safety and preserves Adam's continuous-time dissipative dynamics. Empirically, AdamO is broadly compatible with diverse offline RL baselines, improving stability and returns across a broad suite of benchmarks.

cs.LG↗

arXiv2026

Cloud-Edge Collaborative Large Models for Robust Photovoltaic Power Forecasting

Photovoltaic (PV) power forecasting in edge-enabled grids requires balancing forecasting accuracy, robustness under weather-driven distribution shifts, and strict latency constraints. Existing models work well under normal conditions but often struggle with rare ramp events and unexpected weather changes. Relying solely on cloud-based large models often leads to significant communication delays, which can hinder timely and efficient forecasting in practical grid environments. To address these issues, we propose a condition-adaptive cloud-edge collaborative framework *CAPE* for PV forecasting. *CAPE* consists of three main modules: a site-specific expert model for routine predictions, a lightweight edge-side model for enhanced local inference, and a cloud-based large retrieval model that provides relevant historical cases when needed. These modules are coordinated by a screening module that evaluates uncertainty, out-of-distribution risk, weather mutations, and model disagreement. Furthermore, we employ a Lyapunov-guided routing strategy to dynamically determine when to escalate inference to more powerful models under long-term system constraints. The final forecast is produced through adaptive fusion of the selected model outputs. Experiments on two real-world PV datasets demonstrate that *CAPE* achieves superior performance in terms of forecasting accuracy, robustness, routing quality, and system efficiency.

cs.LG↗

arXiv2026

FORLER: Federated Offline Reinforcement Learning with Q-Ensemble and Actor Rectification

In Internet-of-Things systems, federated learning has advanced online reinforcement learning (RL) by enabling parallel policy training without sharing raw data. However, interacting with real environments online can be risky and costly, motivating offline federated RL (FRL), where local devices learn from fixed datasets. Despite its promise, offline FRL may break down under low-quality, heterogeneous data. Offline RL tends to get stuck in local optima, and in FRL, one device's suboptimal policy can degrade the aggregated model, i.e., policy pollution. We present FORLER, combining Q-ensemble aggregation on the server with actor rectification on devices. The server robustly merges device Q-functions to curb policy pollution and shift heavy computation off resource-constrained hardware without compromising privacy. Locally, actor rectification enriches policy gradients via a zeroth-order search for high-Q actions plus a bespoke regularizer that nudges the policy toward them. A $δ$-periodic strategy further reduces local computation. We theoretically provide safe policy improvement performance guarantees. Extensive experiments show FORLER consistently outperforms strong baselines under varying data quality and heterogeneity.

cs.LG↗

arXiv2026

Less is More: Clustered Cross-Covariance Control for Offline RL

A fundamental challenge in offline reinforcement learning is distributional shift. Scarce data or datasets dominated by out-of-distribution (OOD) areas exacerbate this issue. Our theoretical analysis and experiments show that the standard squared error objective induces a harmful TD cross covariance. This effect amplifies in OOD areas, biasing optimization and degrading policy learning. To counteract this mechanism, we develop two complementary strategies: partitioned buffer sampling that restricts updates to localized replay partitions, attenuates irregular covariance effects, and aligns update directions, yielding a scheme that is easy to integrate with existing implementations, namely Clustered Cross-Covariance Control for TD (C^4). We also introduce an explicit gradient-based corrective penalty that cancels the covariance induced bias within each update. We prove that buffer partitioning preserves the lower bound property of the maximization objective, and that these constraints mitigate excessive conservatism in extreme OOD areas without altering the core behavior of policy constrained offline reinforcement learning. Empirically, our method showcases higher stability and up to 30% improvement in returns over prior methods, especially with small datasets and splits that emphasize OOD areas.

cs.LG↗

arXiv2025

FOVA: Offline Federated Reinforcement Learning with Mixed-Quality Data

Offline Federated Reinforcement Learning (FRL), a marriage of federated learning and offline reinforcement learning, has attracted increasing interest recently. Albeit with some advancement, we find that the performance of most existing offline FRL methods drops dramatically when provided with mixed-quality data, that is, the logging behaviors (offline data) are collected by policies with varying qualities across clients. To overcome this limitation, this paper introduces a new vote-based offline FRL framework, named FOVA. It exploits a \emph{vote mechanism} to identify high-return actions during local policy evaluation, alleviating the negative effect of low-quality behaviors from diverse local learning policies. Besides, building on advantage-weighted regression (AWR), we construct consistent local and global training objectives, significantly enhancing the efficiency and stability of FOVA. Further, we conduct an extensive theoretical analysis and rigorously show that the policy learned by FOVA enjoys strict policy improvement over the behavioral policy. Extensive experiments corroborate the significant performance gains of our proposed algorithm over existing baselines on widely used benchmarks.

cs.LG↗

arXiv2025

Scalable and Distributed Individualized Treatment Rules for Massive Datasets

Synthesizing information from multiple data sources is crucial for constructing accurate individualized treatment rules (ITRs). However, privacy concerns often present significant barriers to the integrative analysis of such multi-source data. Classical meta-learning, which averages local estimates to derive the final ITR, is frequently suboptimal due to biases in these local estimates. To address these challenges, we propose a convolution-smoothed weighted support vector machine for learning the optimal ITR. The accompanying loss function is both convex and smooth, which allows us to develop an efficient multi-round distributed learning procedure for ITRs. Such distributed learning ensures optimal statistical performance with a fixed number of communication rounds, thereby minimizing coordination costs across data centers while preserving data privacy. Our method avoids pooling subject-level raw data and instead requires only sharing summary statistics. Additionally, we develop an efficient coordinate gradient descent algorithm, which guarantees at least linear convergence for the resulting optimization problem. Extensive simulations and an application to sepsis treatment across multiple intensive care units validate the effectiveness of the proposed method.

stat.ME↗

arXiv2025

Details Matter for Indoor Open-vocabulary 3D Instance Segmentation

Unlike closed-vocabulary 3D instance segmentation that is often trained end-to-end, open-vocabulary 3D instance segmentation (OV-3DIS) often leverages vision-language models (VLMs) to generate 3D instance proposals and classify them. While various concepts have been proposed from existing research, we observe that these individual concepts are not mutually exclusive but complementary. In this paper, we propose a new state-of-the-art solution for OV-3DIS by carefully designing a recipe to combine the concepts together and refining them to address key challenges. Our solution follows the two-stage scheme: 3D proposal generation and instance classification. We employ robust 3D tracking-based proposal aggregation to generate 3D proposals and remove overlapped or partial proposals by iterative merging/removal. For the classification stage, we replace the standard CLIP model with Alpha-CLIP, which incorporates object masks as an alpha channel to reduce background noise and obtain object-centric representation. Additionally, we introduce the standardized maximum similarity (SMS) score to normalize text-to-proposal similarity, effectively filtering out false positives and boosting precision. Our framework achieves state-of-the-art performance on ScanNet200 and S3DIS across all AP and AR metrics, even surpassing an end-to-end closed-vocabulary method.

cs.CV↗

arXiv2025

Rank-based transfer learning for high-dimensional survival data with application to sepsis data

Sepsis remains a critical challenge due to its high mortality and complex prognosis. To address data limitations in studying MSSA sepsis, we extend existing transfer learning frameworks to accommodate transformation models for high-dimensional survival data. Specifically, we construct a measurement index based on C-index for intelligently identifying the helpful source datasets, and the target model performance is improved by leveraging information from the identified source datasets via performing the transfer step and debiasing step. We further provide an algorithm to construct confidence intervals for each coefficient component. Another significant development is that statistical properties are rigorously established, including $\ell_1/\ell_2$-estimation error bounds of the transfer learning algorithm, detection consistency property of the transferable source detection algorithm and asymptotic theories for the confidence interval construction. Extensive simulations and analysis of MIMIC-IV sepsis data demonstrate the estimation and prediction accuracy, and practical advantages of our approach, providing significant improvements in survival estimates for MSSA sepsis patients.

stat.AP↗

arXiv2025

Efficient Distributed Learning over Decentralized Networks with Convoluted Support Vector Machine

This paper addresses the problem of efficiently classifying high-dimensional data over decentralized networks. Penalized support vector machines (SVMs) are widely used for high-dimensional classification tasks. However, the double nonsmoothness of the objective function poses significant challenges in developing efficient decentralized learning methods. Many existing procedures suffer from slow, sublinear convergence rates. To overcome this limitation, we consider a convolution-based smoothing technique for the nonsmooth hinge loss function. The resulting loss function remains convex and smooth. We then develop an efficient generalized alternating direction method of multipliers (ADMM) algorithm for solving penalized SVM over decentralized networks. Our theoretical contributions are twofold. First, we establish that our generalized ADMM algorithm achieves provable linear convergence with a simple implementation. Second, after a sufficient number of ADMM iterations, the final sparse estimator attains near-optimal statistical convergence and accurately recovers the true support of the underlying parameters. Extensive numerical experiments on both simulated and real-world datasets validate our theoretical findings.

stat.ML↗

arXiv2024

Efficient molecular conformation generation with quantum-inspired algorithm

Conformation generation, also known as molecular unfolding (MU), is a crucial step in structure-based drug design, remaining a challenging combinatorial optimization problem. Quantum annealing (QA) has shown great potential for solving certain combinatorial optimization problems over traditional classical methods such as simulated annealing (SA). However, a recent study showed that a 2000-qubit QA hardware was still unable to outperform SA for the MU problem. Here, we propose the use of quantum-inspired algorithm to solve the MU problem, in order to go beyond traditional SA. We introduce a highly-compact phase encoding method which can exponentially reduce the representation space, compared with the previous one-hot encoding method. For benchmarking, we tested this new approach on the public QM9 dataset generated by density functional theory (DFT). The root-mean-square deviation between the conformation determined by our approach and DFT is negligible (less than about 0.5 Angstrom), which underpins the validity of our approach. Furthermore, the median time-to-target metric can be reduced by a factor of five compared to SA. Additionally, we demonstrate a simulation experiment by MindQuantum using quantum approximate optimization algorithm (QAOA) to reach optimal results. These results indicate that quantum-inspired algorithms can be applied to solve practical problems even before quantum hardware become mature.

quant-ph↗

arXiv2024

Quantum molecular docking with quantum-inspired algorithm

Molecular docking (MD) is a crucial task in drug design, which predicts the position, orientation, and conformation of the ligand when bound to a target protein. It can be interpreted as a combinatorial optimization problem, where quantum annealing (QA) has shown promising advantage for solving combinatorial optimization. In this work, we propose a novel quantum molecular docking (QMD) approach based on QA-inspired algorithm. We construct two binary encoding methods to efficiently discretize the degrees of freedom with exponentially reduced number of bits and propose a smoothing filter to rescale the rugged objective function. We propose a new quantum-inspired algorithm, hopscotch simulated bifurcation (hSB), showing great advantage in optimizing over extremely rugged energy landscapes. This hSB can be applied to any formulation of objective function under binary variables. An adaptive local continuous search is also introduced for further optimization of the discretized solution from hSB. Concerning the stability of docking, we propose a perturbation detection method to help ranking the candidate poses. We demonstrate our approach on a typical dataset. QMD has shown advantages over the search-based Autodock Vina and the deep-learning DIFFDOCK in both re-docking and self-docking scenarios. These results indicate that quantum-inspired algorithms can be applied to solve practical problems in the drug discovery even before quantum hardware become mature.

physics.chem-ph↗

arXiv2024

Quantum-Inspired Machine Learning for Molecular Docking

Molecular docking is an important tool for structure-based drug design, accelerating the efficiency of drug development. Complex and dynamic binding processes between proteins and small molecules require searching and sampling over a wide spatial range. Traditional docking by searching for possible binding sites and conformations is computationally complex and results poorly under blind docking. Quantum-inspired algorithms combining quantum properties and annealing show great advantages in solving combinatorial optimization problems. Inspired by this, we achieve an improved in blind docking by using quantum-inspired combined with gradients learned by deep learning in the encoded molecular space. Numerical simulation shows that our method outperforms traditional docking algorithms and deep learning-based algorithms over 10\%. Compared to the current state-of-the-art deep learning-based docking algorithm DiffDock, the success rate of Top-1 (RMSD<2) achieves an improvement from 33\% to 35\% in our same setup. In particular, a 6\% improvement is realized in the high-precision region(RMSD<1) on molecules data unseen in DiffDock, which demonstrates the well-generalized of our method.

physics.chem-ph↗

arXiv2023

PoPeC: PAoI-Centric Task Offloading with Priority over Unreliable Channels

Freshness-aware computation offloading has garnered great attention recently in the edge computing arena, with the aim of promptly obtaining up-to-date information and minimizing the transmission of outdated data. However, most of the existing work assumes that wireless channels are reliable and neglect the dynamics and stochasticity thereof. In addition, varying priorities of offloading tasks along with heterogeneous computing units also pose significant challenges in effective task scheduling and resource allocation. To address these challenges, we cast the freshness-aware task offloading problem as a multi-priority optimization problem, considering the unreliability of wireless channels, the heterogeneity of edge servers, and prioritized users. Based on the nonlinear fractional programming and ADMM-Consensus method, we propose a joint resource allocation and task offloading algorithm to solve the original problem iteratively. To improve communication efficiency, we further devise a distributed asynchronous variant for the proposed algorithm. We rigorously analyze the performance and convergence of the proposed algorithms and conduct extensive simulations to corroborate their efficacy and superiority over the existing baselines.

cs.NI↗

arXiv2023

ReCLIP: Refine Contrastive Language Image Pre-Training with Source Free Domain Adaptation

Large-scale Pre-Training Vision-Language Model such as CLIP has demonstrated outstanding performance in zero-shot classification, e.g. achieving 76.3% top-1 accuracy on ImageNet without seeing any example, which leads to potential benefits to many tasks that have no labeled data. However, while applying CLIP to a downstream target domain, the presence of visual and text domain gaps and cross-modality misalignment can greatly impact the model performance. To address such challenges, we propose ReCLIP, the first source-free domain adaptation method for vision-language models, which does not require any source data or target labeled data. ReCLIP first learns a projection space to mitigate the misaligned visual-text embeddings and learns pseudo labels, and then deploys cross-modality self-training with the pseudo labels, to update visual and text encoders, refine labels and reduce domain gaps and misalignments iteratively. With extensive experiments, we demonstrate ReCLIP reduces the average error rate of CLIP from 30.17% to 25.06% on 22 image classification benchmarks. Code available at https://github.com/michiganleon/ReCLIP_WACV.

cs.CV↗

arXiv2023

Q-Drug: a Framework to bring Drug Design into Quantum Space using Deep Learning

Optimizing the properties of molecules (materials or drugs) for stronger toughness, lower toxicity, or better bioavailability has been a long-standing challenge. In this context, we propose a molecular optimization framework called Q-Drug (Quantum-inspired optimization algorithm for Drugs) that leverages quantum-inspired algorithms to optimize molecules on discrete binary domain variables. The framework begins by encoding the molecules into binary embeddings using a discrete VAE. The binary embeddings are then used to construct an Ising energy-like objective function, over which the state-of-the-art quantum-inspired optimization algorithm is adopted to find the optima. The binary embeddings corresponding to the optima are decoded to obtain the optimized molecules. We have tested the framework for optimizing drug molecule properties and have found that it outperforms other molecular optimization methods, finding molecules with better properties in 1/20th to 1/10th of the time previously required. The framework can also be deployed directly on various quantum computing equipment, such as laser pulses CIMs, FPGA Ising Machines, and quantum computers based on quantum annealing, among others. Our work demonstrates a new paradigm that leverages the advantages of quantum computing and AI to solve practically useful problems.

quant-ph↗