arXiv subjects

Xingyu Zhang

Publications and source records attributed to Xingyu Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2025

Learning Invariant Causal Mechanism from Vision-Language Models

Contrastive Language-Image Pretraining (CLIP) has achieved remarkable success, but its performance can degrade when fine-tuned in out-of-distribution (OOD) scenarios. We model the prediction process using a Structural Causal Model (SCM) and show that the causal mechanism involving both invariant and variant factors in training environments differs from that in test environments. In contrast, the causal mechanism with solely invariant factors remains consistent across environments. We theoretically prove the existence of a linear mapping from CLIP embeddings to invariant factors, which can be estimated using interventional data. Additionally, we provide a condition to guarantee low OOD risk of the invariant predictor. Based on these insights, we propose the Invariant Causal Mechanism of CLIP (CLIP-ICM) framework. CLIP-ICM involves collecting interventional data, estimating a linear projection matrix, and making predictions within the invariant subspace. Experiments on several OOD datasets show that CLIP-ICM significantly improves the performance of CLIP. Our method offers a simple but powerful enhancement, boosting the reliability of CLIP in real-world applications.

cs.CV↗

arXiv2025

SLOT: Sample-specific Language Model Optimization at Test-time

We propose SLOT (Sample-specific Language Model Optimization at Test-time), a novel and parameter-efficient test-time inference approach that enhances a language model's ability to more accurately respond to individual prompts. Existing Large Language Models (LLMs) often struggle with complex instructions, leading to poor performances on those not well represented among general samples. To address this, SLOT conducts few optimization steps at test-time to update a light-weight sample-specific parameter vector. It is added to the final hidden layer before the output head, and enables efficient adaptation by caching the last layer features during per-sample optimization. By minimizing the cross-entropy loss on the input prompt only, SLOT helps the model better aligned with and follow each given instruction. In experiments, we demonstrate that our method outperforms the compared models across multiple benchmarks and LLMs. For example, Qwen2.5-7B with SLOT achieves an accuracy gain of 8.6% on GSM8K from 57.54% to 66.19%, while DeepSeek-R1-Distill-Llama-70B with SLOT achieves a SOTA accuracy of 68.69% on GPQA among 70B-level models. Our code is available at https://github.com/maple-research-lab/SLOT.

cs.CL↗

arXiv2025

Don't Shake the Wheel: Momentum-Aware Planning in End-to-End Autonomous Driving

End-to-end autonomous driving frameworks enable seamless integration of perception and planning but often rely on one-shot trajectory prediction, which may lead to unstable control and vulnerability to occlusions in single-frame perception. To address this, we propose the Momentum-Aware Driving (MomAD) framework, which introduces trajectory momentum and perception momentum to stabilize and refine trajectory predictions. MomAD comprises two core components: (1) Topological Trajectory Matching (TTM) employs Hausdorff Distance to select the optimal planning query that aligns with prior paths to ensure coherence;(2) Momentum Planning Interactor (MPI) cross-attends the selected planning query with historical queries to expand static and dynamic perception files. This enriched query, in turn, helps regenerate long-horizon trajectory and reduce collision risks. To mitigate noise arising from dynamic environments and detection errors, we introduce robust instance denoising during training, enabling the planning model to focus on critical signals and improve its robustness. We also propose a novel Trajectory Prediction Consistency (TPC) metric to quantitatively assess planning stability. Experiments on the nuScenes dataset demonstrate that MomAD achieves superior long-term consistency (>=3s) compared to SOTA methods. Moreover, evaluations on the curated Turning-nuScenes shows that MomAD reduces the collision rate by 26% and improves TPC by 0.97m (33.45%) over a 6s prediction horizon, while closedloop on Bench2Drive demonstrates an up to 16.3% improvement in success rate.

cs.RO↗

arXiv2025

AI-Driven Healthcare: A Review on Ensuring Fairness and Mitigating Bias

Artificial intelligence (AI) is rapidly advancing in healthcare, enhancing the efficiency and effectiveness of services across various specialties, including cardiology, ophthalmology, dermatology, emergency medicine, etc. AI applications have significantly improved diagnostic accuracy, treatment personalization, and patient outcome predictions by leveraging technologies such as machine learning, neural networks, and natural language processing. However, these advancements also introduce substantial ethical and fairness challenges, particularly related to biases in data and algorithms. These biases can lead to disparities in healthcare delivery, affecting diagnostic accuracy and treatment outcomes across different demographic groups. This review paper examines the integration of AI in healthcare, highlighting critical challenges related to bias and exploring strategies for mitigation. We emphasize the necessity of diverse datasets, fairness-aware algorithms, and regulatory frameworks to ensure equitable healthcare delivery. The paper concludes with recommendations for future research, advocating for interdisciplinary approaches, transparency in AI decision-making, and the development of innovative and inclusive AI applications.

cs.AI↗

arXiv2025

Photon-number-resolving single-photon detector with a system detection efficiency of 98% and photon-number resolution of 32

Efficiently distinguishing photon numbers is a crucial yet challenging technology for various quantum information and quantum metrology applications. While superconducting transition edge sensors offer good photon-number-resolving (PNR) capabilities, they are hampered by low detection speed, timing jitter, and complex cooling and readout requirements. In this work, we present a significant advancement toward achieving high-fidelity PNR single-photon detectors. The unique twin-layer configuration of superconducting nanowire atop a dielectric mirror ensures the near-unity detection efficiency. The segmented design enables spatial multiplexing, establishing a mapping relationship between pulse amplitude and registered photons. The fabricated detector exhibits impressive performance metrics, including a single-photon system detection efficiency (SDE) of ~ 98% at a dark count rate of 20 cps and photon-number resolution capability up to 32. Further characterization through detector tomography reveals high fidelities for two-, three-, and four-photon events, approximately 87%,73%, and 40% respectively. Moreover, the detector operates at a high count rate of 41 MHz at 3dB-SDE, with a low timing jitter of as low as 40 ps. With its near-unity efficiency, high photon-number resolution, low dark count rate and fast detection speed, we expect significant interest in these detectors, promising substantial benefits for weak light detection and optical quantum information applications.

quant-ph↗

arXiv2025

OccRWKV: Rethinking Efficient 3D Semantic Occupancy Prediction with Linear Complexity

3D semantic occupancy prediction networks have demonstrated remarkable capabilities in reconstructing the geometric and semantic structure of 3D scenes, providing crucial information for robot navigation and autonomous driving systems. However, due to their large overhead from dense network structure designs, existing networks face challenges balancing accuracy and latency. In this paper, we introduce OccRWKV, an efficient semantic occupancy network inspired by Receptance Weighted Key Value (RWKV). OccRWKV separates semantics, occupancy prediction, and feature fusion into distinct branches, each incorporating Sem-RWKV and Geo-RWKV blocks. These blocks are designed to capture long-range dependencies, enabling the network to learn domain-specific representation (i.e., semantics and geometry), which enhances prediction accuracy. Leveraging the sparse nature of real-world 3D occupancy, we reduce computational overhead by projecting features into the bird's-eye view (BEV) space and propose a BEV-RWKV block for efficient feature enhancement and fusion. This enables real-time inference at 22.2 FPS without compromising performance. Experiments demonstrate that OccRWKV outperforms the state-of-the-art methods on the SemanticKITTI dataset, achieving a mIoU of 25.1 while being 20 times faster than the best baseline, Co-Occ, making it suitable for real-time deployment on robots to enhance autonomous navigation efficiency. Code and video are available on our project page: https://jmwang0117.github.io/OccRWKV/.

cs.CV↗

arXiv2025

LipGen: Viseme-Guided Lip Video Generation for Enhancing Visual Speech Recognition

Visual speech recognition (VSR), commonly known as lip reading, has garnered significant attention due to its wide-ranging practical applications. The advent of deep learning techniques and advancements in hardware capabilities have significantly enhanced the performance of lip reading models. Despite these advancements, existing datasets predominantly feature stable video recordings with limited variability in lip movements. This limitation results in models that are highly sensitive to variations encountered in real-world scenarios. To address this issue, we propose a novel framework, LipGen, which aims to improve model robustness by leveraging speech-driven synthetic visual data, thereby mitigating the constraints of current datasets. Additionally, we introduce an auxiliary task that incorporates viseme classification alongside attention mechanisms. This approach facilitates the efficient integration of temporal information, directing the model's focus toward the relevant segments of speech, thereby enhancing discriminative capabilities. Our method demonstrates superior performance compared to the current state-of-the-art on the lip reading in the wild (LRW) dataset and exhibits even more pronounced advantages under challenging conditions.

cs.CV↗

arXiv2024

The GD-1 stellar stream perturber as a core-collapsed self-interacting dark matter halo

The GD-1 stellar stream exhibits spur and gap structures that may result from a close encounter with a dense substructure. When interpreted as a dark matter subhalo, the perturber is denser than predicted in the standard cold dark matter (CDM) model. In self-interacting dark matter (SIDM), however, a halo could evolve into a phase of gravothermal collapse, resulting in a higher central density than its CDM counterpart. We conduct high-resolution controlled N-body simulations to show that a collapsed SIDM halo could account for the GD-1 perturber's high density. We model a progenitor halo with a mass of $3\times10^8~M_\odot$, motivated by a cosmological simulation of a Milky Way analog, and evolve it in the Milky Way's tidal field. For a cross section per mass of $σ/m\approx30-100~{\rm cm^2~g^{-1}}$ at $V_{\rm max }\sim10~{\rm km~s^{-1}}$, the enclosed mass of the SIDM halo within the inner $10~{\rm pc}$ can be increased by more than an order of magnitude compared to its CDM counterpart, leading to a good agreement with the properties of the GD-1 perturber. Our findings indicate that stellar streams provide a novel probe into the self-interacting nature of dark matter.

astro-ph.GA↗

arXiv2024

DRAL: Deep Reinforcement Adaptive Learning for Multi-UAVs Navigation in Unknown Indoor Environment

Autonomous indoor navigation of UAVs presents numerous challenges, primarily due to the limited precision of GPS in enclosed environments. Additionally, UAVs' limited capacity to carry heavy or power-intensive sensors, such as overheight packages, exacerbates the difficulty of achieving autonomous navigation indoors. This paper introduces an advanced system in which a drone autonomously navigates indoor spaces to locate a specific target, such as an unknown Amazon package, using only a single camera. Employing a deep learning approach, a deep reinforcement adaptive learning algorithm is trained to develop a control strategy that emulates the decision-making process of an expert pilot. We demonstrate the efficacy of our system through real-time simulations conducted in various indoor settings. We apply multiple visualization techniques to gain deeper insights into our trained network. Furthermore, we extend our approach to include an adaptive control algorithm for coordinating multiple drones to lift an object in an indoor environment collaboratively. Integrating our DRAL algorithm enables multiple UAVs to learn optimal control strategies that adapt to dynamic conditions and uncertainties. This innovation enhances the robustness and flexibility of indoor navigation and opens new possibilities for complex multi-drone operations in confined spaces. The proposed framework highlights significant advancements in adaptive control and deep reinforcement learning, offering robust solutions for complex multi-agent systems in real-world applications.

cs.RO↗

arXiv2024

Optimized Coordination Strategy for Multi-Aerospace Systems in Pick-and-Place Tasks By Deep Neural Network

In this paper, we present an advanced strategy for the coordinated control of a multi-agent aerospace system, utilizing Deep Neural Networks (DNNs) within a reinforcement learning framework. Our approach centers on optimizing autonomous task assignment to enhance the system's operational efficiency in object relocation tasks, framed as an aerospace-oriented pick-and-place scenario. By modeling this coordination challenge within a MuJoCo environment, we employ a deep reinforcement learning algorithm to train a DNN-based policy to maximize task completion rates across the multi-agent system. The objective function is explicitly designed to maximize effective object transfer rates, leveraging neural network capabilities to handle complex state and action spaces in high-dimensional aerospace environments. Through extensive simulation, we benchmark the proposed method against a heuristic combinatorial approach rooted in game-theoretic principles, demonstrating a marked performance improvement, with the trained policy achieving up to 16\% higher task efficiency. Experimental validation is conducted on a multi-agent hardware setup to substantiate the efficacy of our approach in a real-world aerospace scenario.

cs.RO↗

arXiv2024

Optimally Fast Qubit Reset

In practice, qubit reset must be operated in an extremely short time, which incurs a thermodynamic cost within multiple orders of magnitude above the Landauer bound. We present a general framework to determine the minimal thermodynamic cost and the optimal protocol for arbitrary resetting speeds. Our study reveals the divergent behavior of minimal entropy production in the short-time limit depends on the convergence and divergence of the jump operators. For the convergent class, an inherent trade-off exists between the minimal required time and the set error probability, which hinders the Moore's law continuing in such cases. Moreover, we find the optimal protocol exhibits the similarity in the fast-driving regime for different times. To demonstrate our findings, we empoly fermionic and bosonic baths as examples. Our results suggest that the super-Ohmic bosonic heat bath is a suitable choice for qubit reset.

cond-mat.stat-mech↗

arXiv2024

Not All Frequencies Are Created Equal:Towards a Dynamic Fusion of Frequencies in Time-Series Forecasting

Long-term time series forecasting is a long-standing challenge in various applications. A central issue in time series forecasting is that methods should expressively capture long-term dependency. Furthermore, time series forecasting methods should be flexible when applied to different scenarios. Although Fourier analysis offers an alternative to effectively capture reusable and periodic patterns to achieve long-term forecasting in different scenarios, existing methods often assume high-frequency components represent noise and should be discarded in time series forecasting. However, we conduct a series of motivation experiments and discover that the role of certain frequencies varies depending on the scenarios. In some scenarios, removing high-frequency components from the original time series can improve the forecasting performance, while in others scenarios, removing them is harmful to forecasting performance. Therefore, it is necessary to treat the frequencies differently according to specific scenarios. To achieve this, we first reformulate the time series forecasting problem as learning a transfer function of each frequency in the Fourier domain. Further, we design Frequency Dynamic Fusion (FreDF), which individually predicts each Fourier component, and dynamically fuses the output of different frequencies. Moreover, we provide a novel insight into the generalization ability of time series forecasting and propose the generalization bound of time series forecasting. Then we prove FreDF has a lower bound, indicating that FreDF has better generalization ability. Extensive experiments conducted on multiple benchmark datasets and ablation studies demonstrate the effectiveness of FreDF. The code is available at https://github.com/Zh-XY22/FreDF.

cs.LG↗

arXiv2024

Canonical-Polyadic-Decomposition of the Potential Energy Surface Fitted by Warm-Started Support Vector Regression

In this work, we propose a decoupled support vector regression (SVR) approach for direct canonical polyadic decomposition (CPD) of a potential energy surface (PES) through a set of discrete training energy data. This approach, denoted by CPD-SVR, is able to directly construct the PES in CPD with a more compressed form than previously developed Gaussian process regression (GPR) for CPD, denoted by CPD-GRP ({\it J. Phys. Chem. Lett.} {\bf 13} (2022), 11128). Similar to CPD-GPR, the present CPD-SVR method requires the multi-dimension kernel function in a product of a series of one-dimensional functions. We shall show that, only a small set of support vectors play a role in SVR prediction making CPD-SVR predict lower-rank CPD than CPD-GPR. To save computational cost in determining support vectors, we propose a warm-started (ws) algorithm where a pre-existed crude PES is employed to classify the training data. With the warm-started algorithm, the present CPD-SVR approach is extended to the CPD-ws-SVR approach. Then, we test CPD-ws-SVR and compare it with CPD-GPR through constructions and applications of the PESs of H + H$_2$, H$_2$ + H$_2$, and H$_2$/Cu(111). To this end, the training data are computed by existed PESs. Calculations on H + H$_2$ predict a good agreement of dynamics results among various CPD forms, which are constructed through different approaches.

physics.chem-ph↗

arXiv2024

Senna: Bridging Large Vision-Language Models and End-to-End Autonomous Driving

End-to-end autonomous driving demonstrates strong planning capabilities with large-scale data but still struggles in complex, rare scenarios due to limited commonsense. In contrast, Large Vision-Language Models (LVLMs) excel in scene understanding and reasoning. The path forward lies in merging the strengths of both approaches. Previous methods using LVLMs to predict trajectories or control signals yield suboptimal results, as LVLMs are not well-suited for precise numerical predictions. This paper presents Senna, an autonomous driving system combining an LVLM (Senna-VLM) with an end-to-end model (Senna-E2E). Senna decouples high-level planning from low-level trajectory prediction. Senna-VLM generates planning decisions in natural language, while Senna-E2E predicts precise trajectories. Senna-VLM utilizes a multi-image encoding approach and multi-view prompts for efficient scene understanding. Besides, we introduce planning-oriented QAs alongside a three-stage training strategy, which enhances Senna-VLM's planning performance while preserving commonsense. Extensive experiments on two datasets show that Senna achieves state-of-the-art planning performance. Notably, with pre-training on a large-scale dataset DriveX and fine-tuning on nuScenes, Senna significantly reduces average planning error by 27.12% and collision rate by 33.33% over model without pre-training. We believe Senna's cross-scenario generalization and transferability are essential for achieving fully autonomous driving. Code and models will be released at https://github.com/hustvl/Senna.

cs.CV↗

arXiv2024

Drone based superconducting single photon detection system with detection efficiency more than 90%

Bounded by the size, weight, and power consumption (SWaP) of conventional superconducting single photon detectors (SSPD), applications of SSPDs were commonly confined in the laboratory. However, booming demands for high efficiency single photon detector incorporated with avionic platforms arise with the development of remote imaging and sensing or long-haul quantum communication without topographical constraints. We herein designed and manufactured the first drone based SSPD system with a SDE as high as 91.8%. This drone based SSPD system is established with high performance NbTiN SSPDs, self-developed miniature liquid helium dewar, and homemade integrated electric setups, which is able to be launched in complex topographical conditions. Such a drone based SSPD system may open the use of SSPDs for applications that demand high-SDE in complex environments.

cond-mat.supr-con↗

arXiv2024

Self-Adaptive Robust Motion Planning for High DoF Robot Manipulator using Deep MPC

In contemporary control theory, self-adaptive methodologies are highly esteemed for their inherent flexibility and robustness in managing modeling uncertainties. Particularly, robust adaptive control stands out owing to its potent capability of leveraging robust optimization algorithms to approximate cost functions and relax the stringent constraints often associated with conventional self-adaptive control paradigms. Deep learning methods, characterized by their extensive layered architecture, offer significantly enhanced approximation prowess. Notwithstanding, the implementation of deep learning is replete with challenges, particularly the phenomena of vanishing and exploding gradients encountered during the training process. This paper introduces a self-adaptive control scheme integrating a deep MPC, governed by an innovative weight update law designed to mitigate the vanishing and exploding gradient predicament by employing the gradient sign exclusively. The proffered controller is a self-adaptive dynamic inversion mechanism, integrating an augmented state observer within an auxiliary estimation circuit to enhance the training phase. This approach enables the deep MPC to learn the entire plant model in real-time and the efficacy of the controller is demonstrated through simulations involving a high-DoF robot manipulator, wherein the controller adeptly learns the nonlinear plant dynamics expeditiously and exhibits commendable performance in the motion planning task.

cs.RO↗

arXiv2024

Collective advantages in qubit reset: effect of coherent qubits

The Landauer principle sets a lower bound on the thermodynamic cost of qubit reset, which is only attainable for the quasistatic process. In this Letter, we explore the collective advantage of qubit reset of coherent qubits in three aspects. First, for the quasistatic process, the thermodynamic cost of collective reset is remarkably lower than parallel reset because of the reduced Hilbert space dimension due to entanglement effects. Second, for the finite-time qubit reset, we prove that the error probability fades away and per-qubit heat production tends the Landauer bound for initially continuous protocols in the thermodynamic limit. Third, we show that qubit reset performance enhances with the increase in the number of qubits. Our results, illustrated by different protocols, provide a blueprint for future quantum device fabrication.

quant-ph↗

arXiv2024

Self-interacting dark matter interpretation of Crater II

The satellite galaxy Crater II of the Milky Way is extremely cold and exceptionally diffuse. These unusual properties are challenging to understand in the standard model of cold dark matter. We investigate the formation of Crater II in self-interacting dark matter (SIDM), where dark matter particles can scatter and thermalize. We conduct a series of controlled N-body simulations to model the tidal evolution of Crater II, varying the self-interacting cross section, orbit parameters, and initial stellar distribution. Dark matter self-interactions lead to halo core formation and the distribution of stars expands accordingly. A cored SIDM halo also boosts tidal mass loss, allowing for a high orbit. Our simulations show that SIDM halos with a $1~{\rm kpc}$ core can simultaneously explain the low stellar velocity dispersion and the large half-light radius of Crater II, remaining robust to the initial distribution of stars. For the orbit motivated by the measurements from Gaia Early Data Release 3, the favored self-interacting cross section is approximately $60~{\rm cm^2/g}$ on the mass scale of Crater II.

astro-ph.GA↗