Searcharxiv⌕ Search

arXiv subjects

Feifei Gao

Publications and source records attributed to Feifei Gao.

At least 19 recordsLinked to original sources

HandRSI: Recursive Self-Improvement for Hand Pose Estimation

Egocentric 3D hand pose estimators still miss hands at the image border, misplace occluded hands and jitter. Such failures are usually fixed by hand-written inference refinements and by fine-tuning, two interdependent steps that are tuned separately. We present HandRSI, a recursive self-improvement framework for a hand pose system that consists of a video estimator and its executable inference harness. In each round, coding agents refine the harness, and a change is accepted only if it improves a composite of all ten benchmark metrics on held-out data without significantly regressing any of them. The accepted harness then guides the estimator through selective distillation, which supervises hand presence and temporal consistency on frames without ground truth, while ground truth supervises pose. Since the updated estimator may change which refinements remain useful, the harness is re-validated on it. Starting from the released ACE-Ego-Hand model, five rounds reduce MPJPE-p by 35% on HOT3D and 11% on ARCTIC and roughly halve jitter. On HOT3D, HandRSI's MPJPE-p is 16% lower than that of ACE-Ego-Hand fine-tuned on the same ground truth. Under a unified protocol, it outperforms nine re-run published methods on all ten metrics of both benchmarks. Hypervolume selection on the accuracy-speed front yields Lite, Medium and Heavy systems running at 20.7 to 41.1 FPS. This report gives the full method, every stage of every round and the complete protocol. Code, evaluator, harnesses and weights are public.

cs.CV↗

HarnessMM: Recursive Self-Improvement for Harness Token Efficiency

HarnessMinusMinus (HarnessMM) is a research project on improving the efficiency of agent harnesses, the software coordinating model reasoning, tool execution, and feedback. This technical report consolidates its shared framework, implementations, and empirical studies across coding, tool-based, and visual-agent settings. Agent harnesses often invoke reasoning after each tool result, even when earlier calls have already determined subsequent actions. HarnessMM addresses this overhead through guarded decision reuse and recursive self-improvement (RSI). RSI refines author prompts, context settings, and adapter rules using training feedback, with validation selecting deployment configurations. At inference time, a controller reuses resolved decisions while keeping actor weights fixed. Executable Decision Graphs encode actions, arguments, dependencies, and guards. Guarded Decision Reuse activates eligible decisions from observed feedback, while Certified Callback Substitution executes them through native tool interfaces, replacing redundant reasoning callbacks. The original actor resumes when new reasoning is required or reuse conditions are unmet. On HumanEval, HarnessMM preserves Codex success at 90.82\% while reducing tokens per task from 66.72k to 47.02k (29.54\% saving). For DeepSeek Harness, success improves from 71.43\% to 85.71\% while tokens decrease from 96.28k to 11.62k (87.93\% saving). An MBPP ablation shows 26.91\% lower cost than single-use graphs and 19.52\% lower cost than retained callbacks. The report also summarizes the HarnessMM-V visual-agent study and documents optimization overhead and execution time where measured. Together, these studies investigate guarded decision reuse across diverse agent harness settings. Code is available at https://github.com/YuhuaJiang2002/HarnessMM.

cs.AI↗

Mini-Batch Stochastic Conjugate Gradient Algorithms with Minimal Variance

Stochastic gradient descent (SGD) methods have gained widespread popularity for solving large-scale optimization problems. However, the inherent variance in SGD often leads to slow convergence rates. We introduce a family of unbiased stochastic gradient estimators that encompasses existing estimators from the literature and identify a gradient estimator that not only maintains unbiasedness but also achieves minimal variance. Compared with the existing estimator used in SGD algorithms, the proposed estimator demonstrates a significant reduction in variance. By utilizing this stochastic gradient estimator to approximate the full gradient, we propose two mini-batch stochastic conjugate gradient algorithms with minimal variance. Under the assumptions of strong convexity and smoothness on the objective function, we prove that the two algorithms achieve linear convergence rates. Numerical experiments validate the effectiveness of the proposed gradient estimator in reducing variance and demonstrate that the two stochastic conjugate gradient algorithms exhibit accelerated convergence rates and enhanced stability.

math.OC↗

Dual Secure Sensing and Communications in MA-Enabled Networked ISAC Systems

Integrated sensing and communication (ISAC) is envisioned as a cornerstone technology for next-generation networks, yet it introduces significant security risks in both communication and sensing domains. To mitigate the coverage limitations inherent in conventional mono-static or bi-static setups, this paper proposes a networked ISAC framework empowered by movable antennas (MAs). By leveraging flexible antenna reconfiguration, MAs provide additional spatial degrees of freedom to significantly boost beamforming gain and interference suppression. Specifically, we investigate a novel dual-domain security paradigm where networked base stations collaboratively protect uplink (UL) and downlink (DL) transmissions from communication eavesdropping while simultaneously safeguarding target sensing data against sensing eavesdroppers (SEs). To this end, we formulate a joint optimization problem to maximize the sum of secrecy rates, subject to sensing requirements, SE eavesdropping probability constraint, and practical MA deployment constraints, by jointly optimizing receive combiners, transmit beamformers, UL power allocation, and MA positions. To address this non-convex problem with highly coupled variables, an efficient alternating optimization algorithm is proposed, integrating closed-form solutions for receive combiners, successive convex approximation for transmit beamforming and UL power, and a tailored sensing-aware multi-candidate particle swarm optimization for the high-dimensional MA positioning. Extensive numerical results validate the effectiveness of the proposed algorithms, demonstrating that the proposed MA-enabled networked cooperative ISAC design significantly outperforms conventional fixed-position antenna-based designs in terms of security performance, while achieving notable power- and antenna-saving gains.

eess.SP↗

Graph Learning with Spectral Connectivity Priors for Scarce Data

Learning a sparse graph from scarce data is practically important but challenging. Motivated by the desirable combination of local sparsity and strong global connectivity exhibited by expander-like graphs, we propose spectral connectivity-regularized graph learning (SCoGL), a framework that incorporates a family of Laplacian spectral priors to explicitly promote global connectivity. Specifically, SCoGL augments a combinatorial-Laplacian-constrained graphical lasso (GLASSO) objective over a target adjacency matrix $\mathbf{W}$ with a general connectivity prior computed from Laplacian eigenvalues. We derive gradients for several representative connectivity priors and develop a projected gradient descent (PGD) algorithm with Armijo backtracking to efficiently optimize $\mathbf{W}$. Experiments show that the proposed SCoGL variants improve graph recovery and enhance downstream tasks such as graph signal denoising when signal observations are scarce.

cs.LG↗

Sampling headroom is not selection gain: a compute-value audit of test-time scaling for video world models

Test-time scaling (TTS) can improve generation only when additional compute produces better candidates and the system can reliably identify them. This distinction is especially important for video world models, where a wider sample pool may contain stronger rollouts without improving the output that is ultimately selected. We introduce the Compute-Value Audit (CVA), a sequential framework that asks whether extra sampling creates opportunity, observable signals provide a reliable state, that state supports a beneficial action, and the resulting gain exceeds the full entry fee of generation and verification. On 192 Physics-IQ scenes, expanding the pool from 4 to 16 candidates increases oracle quality by +9.23 IQ (95% CI [+7.44, +11.14]), but Flow, Cycle, and VideoReward fail to recover this headroom reliably. Across three generators, none of twelve adaptive-depth policies outperforms uniform compute; they recover only 42-69% of the measured entry fee. A matched-60-NFE Predict-and-Perturb intervention on VideoPhy2 is likewise negative across three fresh-seed replicas. These negative results are not universal: anchor-explorer passes all four stages in a sparse PRM800K setting, MMLU-Pro exposes the gap between predictive state and useful action, and a privileged paired future establishes a positive video upper bound. Together, these results show that sampling headroom has deployment value only when it can be converted into a reliable decision whose benefit survives the complete compute charge. Code is available at https://github.com/YuhuaJiang2002/sampling-headroom-is-not-selection-gain.

cs.CV↗

Ultra-High Resolution Method for Multipath Within a Co-Delay-Doppler Bin in DFT-P-OCDM

Communication systems can reuse their transmitted signals for sensing without dedicated radar transmissions. For an established DFT-preprocessed orthogonal chirp division multiplexing (DFT-P-OCDM) waveform, this task becomes difficult when several physical paths in a doubly selective channel fall into the same co-delay-Doppler bin. In this case, the number of resolvable delay classes inferred from the pilot may be smaller than the number of physical paths within the co-bin. This mismatch increases the difficulty of path number determination, fractional Doppler offset estimation, and channel reconstruction. We derive a pointwise relationship between the input and output in the DFT preprocessed Fresnel (DPF) domain for doubly selective channels with multiple paths within the co-delay-Doppler bin. Based on this input and output relation, we propose the two stage ultra high resolution (TSUR) framework. The first stage uses pilots of the phase progression to estimate delay, while the second stage uses the leakage samples to estimate the Doppler and the number of paths within each delay class. Furthermore, we derive CRLBs and analyze how the pilot configuration trades sensing resolution and communication recovery. Simulation results demonstrate that TSUR resolves same delay paths within a co-delay-Doppler bin, remains robust to exist delay offsets, and achieves lower fractional Doppler estimation errors than sequential extraction and off-grid baselines.

cs.IT↗

SCOPE: Score-Isolated Agentic Optimization for Video World Models

Video world models are increasingly used as simulators for planning and embodied decision making, yet improving them at inference time introduces a subtle evaluation problem: prompts, samplers, verifiers, and selectors may evolve together, making it difficult to attribute gains or prevent held-out feedback from shaping the final policy. We introduce \scope (\emph{\scopefullname}), a framework for auditable inference-time adaptation of frozen video world models. \scope represents external controls as a typed state, updates this state only through bounded changes supported by development evidence, and freezes the resulting policy before held-out evaluation. On Physics-IQ benchmark, \scope improves over the exact frozen base by $+14.24$ (95\% CI $[+8.10,+21.23]$). Controlled ablations further identify gains from scene specification, sampling, and learned selection, while the margin over the strongest matched agentic baseline remains unresolved. Cross-backbone and prospective evaluations reveal a complementary result: useful inference-time updates exist, but their benefits do not transfer uniformly across models and settings. Together, these findings suggest that reliable inference-time adaptation requires not only better proposals, but also a principled mechanism for deciding which updates should become part of the deployed system. Code is available at https://github.com/YuhuaJiang2002/SCOPE.

cs.AI↗

New Mid-Band (FR3, 6-24 GHz) XL-MIMO for 6G: Channel Modeling, Algorithm Evaluation, and Field Trials

The new mid-band (FR3, 6-24 GHz) spectrum is expected to play an important role in future 6G networks by providing a favorable balance among coverage, capacity, and deployment feasibility. Meanwhile, extremely large-scale multiple-input multiple-output (XL-MIMO) has emerged as a key enabling technology to exploit the propagation and spatial multiplexing potential of these frequency bands. Firstly, this paper provides a systematic review of spectrum allocation and standardization activities for new mid-band spectrum, together with the 6G spectrum planning strategies of countries and regions. Secondly, the wideband massive MIMO channel sounder is also introduced, which is specially developed for channel measurements of new mid-band with over a thousand elements. Thirdly, propagation characteristics and channel modeling approaches of four representative XL-MIMO architectures, including co-located, cell-free, and intelligent XL-MIMO, are comprehensively reviewed and analyzed, with particular emphasis on near-field propagation, spatial non-stationarity, and capacity performance. Then, recent advances in channel estimation, beamforming, and artificial-intelligence-assisted signal processing are summarized. In addition, the performance of new mid-band XL-MIMO systems equipped with 1536 and 768 antenna elements is comparatively evaluated. Finally, real communication environment prototype system field trials conducted in the Upper 6 GHz (U6GHz) band are used to investigate practical system performance under realistic deployment conditions. The results indicate that the target signal-to-noise ratio is a critical factor affecting XL-MIMO performance in the U6GHz band.

eess.SP↗

Low-Latency Generative Semantic Communication via Channel-Realization Flow Matching

Generative semantic communication receivers deliver high perceptual quality but suffer from prohibitive decoding latency. This bottleneck arises because diffusion receivers rely on stochastic iterative decoding, while existing flow matching receivers employ independent endpoint coupling that ignores the physical source--channel link, yielding unnecessarily long and curved sampling trajectories. In this paper, we reformulate receiver-side recovery as a realization-coupled bridge flow matching problem under explicit bandwidth and power constraints. Specifically, we propose Realization-Coupled Bridge Flow Matching (RC-BFM), where the decoder initializes from a channel-induced semantic state rather than isotropic noise. Crucially, training pairs are linked via a realization-coupled entropic optimal transport (RC-OT) plan that preserves the physical channel realization of each transmission while maintaining robustness to stochastic fading. Furthermore, we identify independent coupling as the fundamental source of a conditional train--test distribution shift in conditional flow matching-based receivers, and derive an end-to-end distortion bound whose discretization error decays as \(O(K^{-2})\). Experiments on CIFAR-10 and FFHQ-64\(\times\)64 over AWGN and Rayleigh fading channels demonstrate that RC-BFM achieves a superior fidelity--perception trade-off, reducing decoding latency by over 10\(\times\) compared to diffusion-based receivers.

cs.IT↗

ISAC and Vision Fusion for Fine-Grained Low-Altitude Target Recognition

In this paper, we propose an integrated sensing and communications (ISAC) and vision fusion framework for fine-grained low-altitude target recognition. Specifically, we first utilize ISAC system to estimate the position of the low-altitude target. Then we adjust the working parameters of the Pan-TiltZoom (PTZ) camera based on the estimated target position, such that the camera can capture the image of tiny low-altitude target from several hundred meters away. After obtaining the wireless echo signal and visual image of low-altitude target, we employ the short-time Fourier transform (STFT) to obtain the micro-Doppler (mD) spectrum of the target from wireless echo signal, and design a conditional generative adversarial network (cGAN)-based denoising network to optimize the quality of the mD spectrum. Meanwhile, we employ YOLOv11 to detect the low-altitude target from visual image, and then crop the smallsized feature image of the target from the original image. Next, we design a fine-grained low-altitude target recognition network with MobileViT, which can fuse the optimized mD spectrum and the cropped feature image to distinguish the subcategory of low-altitude target. Moreover, we generate a joint ISAC and vision dataset (JIVD) for low-altitude target monitoring based on AirSim and Wireless InSite, which includes diverse target subcategories, scenarios, and weather conditions. The effectiveness and superiority of the proposed scheme have been demonstrated by simulation results.

eess.SP↗

Mode Switching for RDARS-Aided ISAC Systems: From Optimization to Deep Unfolding

Reconfigurable distributed antennas and reflecting surface (RDARS) has recently emerged as a promising architecture for integrated sensing and communication (ISAC), owing to its flexible element-wise mode switching between connection and reflection modes. In this paper, to fully reap the benefits of mode configuration, muting elements that can absorb the incident energy are introduced into RDARS-aided ISAC systems to mitigate multi-user interference (MUI) and enhance sensing performance. To draw useful insights, we first investigate the special cases of single-UE communication, single-target sensing, and two-UE communication to reveal the importance of muting elements. Specifically, the maximum communication and sensing signal-to-noise ratio (SNR), and the signal-to-interference-plus-noise ratio (SINR) expressions are respectively derived for the three cases, together with the optimal number of muting elements for explicitly characterizing the tradeoff between reflection gain loss and MUI suppression. Next, we consider the joint waveform and tri-mode switching design for RDARS-aided ISAC systems, where an alternating optimization (AO)-based penalty dual decomposition (APDD) algorithm is proposed to solve the mixed-integer nonlinear programming (MINLP) problem. Furthermore, a model-driven APDD-Net is developed by deeply unfolding the APDD iterations into a layer-wise architecture, where key parameters are learned to reduce the computational complexity and accelerate convergence. Simulation results verify the theoretical findings on the muting gain and demonstrate that the proposed APDD-Net achieves a better tradeoff between communication and sensing performance compared with benchmark schemes.

eess.SP↗

Networked ISAC Enabled Target Recognition Towards Low-Altitude Economy

In this paper, we propose a low-altitude target (LAT) recognition scheme based on multi-base station (BS) collaboration and multi-scale feature fusion for integrated sensing and communications (ISAC) network. Firstly, we formulate the motion equations, echo channels, and echo signals for unmanned aerial vehicle (UAV), bird, vehicle, and pedestrian under multi-BS collaborative monitoring scenario. Then we extract the velocityresolution-preferred time-frequency spectrum, time-resolutionpreferred time-frequency spectrum, and the velocity-transfer time-frequency spectrum observed by each BS from echo signals. We collectively refer to these three types of time-frequency spectrum as the multi-scale feature of the LAT. Next, we design a multi-BS and multi-scale feature fusion enabled LAT recognition network with Swin Transformer, which employs the visualized images of multi-scale feature to jointly recognize the target through deep feature extraction, intra-BS feature interaction, inter-BS feature interaction, and target recognition output. We generate a massive echo signal dataset comprising 1,440,000 samples for LAT recognition within ISAC network. This dataset can serve as a public benchmark to evaluate our proposed scheme and facilitate future research. Simulation results demonstrate that the proposed scheme realizes high recognition accuracy and robust unseen-subtype generalization, confirming the effectiveness of multi-scale feature fusion and the additional gains brought by multi-BS collaboration. The project page is available at: https://alivn999.github.io/COSMOS-Networked-ISAC-Enabl ed-Target-Recognition-Towards-Low-Altitude-Economy/.

eess.SP↗

Wireless large AI model: shaping the AI-empowered future of 6G and beyond

The emergence of sixth-generation and beyond communication systems is expected to fundamentally transform digital experiences through introducing unparalleled levels of intelligence, efficiency, and connectivity. A promising technology poised to enable this revolutionary vision is a wireless large AI model (WLAM), characterized by its exceptional capabilities in data processing, inference, and decision-making. In light of these remarkable capabilities, this paper provides a comprehensive survey of WLAM, explaining its fundamental principles, diverse applications, critical challenges, and future research opportunities. We begin by introducing the background of WLAM and analyzing the key synergies with wireless networks, emphasizing the mutual benefits. Subsequently, we explore the foundational characteristics of WLAM, delving into their unique relevance in wireless environments. Then, the role of WLAM in optimizing wireless communication systems across various use cases and the reciprocal benefits are systematically investigated. Furthermore, we discuss the integration of WLAM with emerging technologies, highlighting their potential to enable transformative capabilities and breakthroughs in wireless communication. Finally, we thoroughly examine the high-level challenges and discuss pivotal future research directions.

cs.IT↗

VLA-GSE: Boosting Parameter-Efficient Fine-Tuning in VLA with Generalized and Specialized Experts

Vision-language-action (VLA) models inherit rich visual-semantic priors from pre-trained vision-language backbones, but adapting them to robotic control remains challenging. Full fine-tuning (FFT) is prone to overfitting on downstream robotic data and catastrophic forgetting of pretrained vision-language capabilities. Parameter-efficient fine-tuning (PEFT) better preserves pre-trained knowledge, yet existing PEFT methods still struggle to adapt effectively to robot control tasks. To address this gap, we propose VLA-GSE, a parameter-efficient VLA fine-tuning framework that improves control adaptation while retaining PEFT's knowledge preservation advantage. Specifically, VLA-GSE (Generalized and Specialized Experts) is initialized by spectrally decomposing the frozen backbone, assigning leading singular components to generalized experts (shared experts) and disjoint residual components to specialized experts (routed experts). This decomposition improves adaptation capacity under a fixed trainable-parameter budget. Under a comparable parameter budget, VLA-GSE updates only 2.51% of the full model parameters and consistently outperforms strong FFT and PEFT baselines. It achieves 81.2% average zero-shot success on LIBERO-Plus, preserves pre-trained VLM capability comparably to LoRA on multimodal understanding benchmarks, and improves real-world manipulation success under multiple distribution shifts. Code is available at: https://github.com/YuhuaJiang2002/VLA-GSE

cs.RO↗

AsyncVLA: Asynchronous Flow Matching for Vision-Language-Action Models

Vision-language-action (VLA) models have recently emerged as a powerful paradigm for building generalist robots. However, traditional VLA models that generate actions through flow matching (FM) typically rely on rigid and uniform time schedules, i.e., synchronous FM (SFM). Without action context awareness and asynchronous self-correction, SFM becomes unstable in long-horizon tasks, where a single action error can cascade into failure. In this work, we propose asynchronous flow matching VLA (AsyncVLA), a novel framework that introduces temporal flexibility in asynchronous FM (AFM) and enables self-correction in action generation. AsyncVLA breaks from the vanilla SFM in VLA models by generating the action tokens in a non-uniform time schedule with action context awareness. Besides, our method introduces the confidence rater to extract confidence of the initially generated actions, enabling the model to selectively refine inaccurate action tokens before execution. Moreover, we propose a unified training procedure for SFM and AFM that endows a single model with both modes, improving KV-cache utilization. Extensive experiments on robotic manipulation benchmarks demonstrate that AsyncVLA is data-efficient and exhibits self-correction ability. AsyncVLA outperforms existing methods across both simulation and real-world evaluations. Our code is available at https://github.com/YuhuaJiang2002/AsyncVLA.

cs.RO↗

CKT-WAM: Parameter-Efficient Context Knowledge Transfer Between World Action Models

World action models (WAMs) provide a powerful generative framework for embodied control, yet transferring knowledge across heterogeneous WAMs remains challenging due to mismatched latent interfaces, high adaptation cost, and the rigidity of conventional distillation objectives. We propose \textbf{CKT-WAM}, a parameter-efficient \textbf{C}ontext \textbf{K}nowledge \textbf{T}ransfer framework that transfers teacher WAM's knowledge into a student WAM through a compact context in the text embedding space, rather than output imitation or dense hidden-state matching. Specifically, CKT-WAM extracts intermediate teacher hidden states, reduces the number of tokens via compressors' learnable-query cross attention (LQCA), and transforms them through an always-on generalized adapter, a lightweight router, and sparsely activated specialized adapters. The resulting context is then appended to the student's conditioning textual embeddings, thereby injecting the transferred knowledge into the student with minimal architectural modification. Experiments show that CKT-WAM consistently improves zero-shot generalization and achieves the best overall performance on LIBERO-Plus, reaching 86.1\% total success rate with only 1.17\% trainable parameters, while approaching full fine-tuning performance. Beyond simulation, CKT-WAM also demonstrates strong real-world long-horizon manipulation ability, achieving the best average success rate of 83.3\% across four multi-step and long-horizon tasks. Code is available at https://github.com/YuhuaJiang2002/CKT-WAM.

cs.RO↗

Nirvana: A Specialized Generalist Model With Task-Aware Memory Mechanism

Large Language Models (LLMs) excel at general language tasks but struggle in specialized domains. Specialized Generalist Models (SGMs) address this by preserving broad capabilities while adapting to target domains. However, existing architectures provide limited support for task-guided specialized memory mechanisms. In this work, we introduce Nirvana, an SGM featuring specialized memory, linear-time complexity, and test-time task information extraction. Central to Nirvana are: (1) Task-Aware Memory Trigger ($\textit{Trigger}$), which treats each input as a self-supervised fine-tuning task and adjusts task-related parameters on the fly; and (2) Specialized Memory Updater ($\textit{Updater}$), which dynamically consolidates task-relevant context. Nirvana matches or surpasses LLM baselines on general benchmarks and achieves the lowest perplexity across specialized domains including biomedicine, finance, and law. On the challenging task of Magnetic Resonance Imaging (MRI), we attach lightweight codecs to the frozen Nirvana backbone and fine-tune them on paired k-space signals and images. Nirvana achieves higher-fidelity reconstructions than conventional LLM-based models, with Trigger providing effective domain-specific adaptation. Ablation studies confirm that removing Trigger leads to substantial degradation across all tasks, underscoring its essential role in task-aware specialization. Models are available at https://huggingface.co/collections/YuhuaJiang/nirvana. Code is available at https://github.com/YuhuaJiang2002/Nirvana.

cs.LG↗