arXiv subjects

Zhiling Wang

Publications and source records attributed to Zhiling Wang.

At least 19 recordsLinked to original sources

arXiv2026

Heralded Leakage Detection with Preserved Computational-State Coherence in a Fixed-Frequency Transmon

Leakage out of the computational subspace is a major error mechanism in superconducting quantum processors. Detecting leakage without disturbing the encoded quantum information can provide a heralded error signal that error-correction decoders exploit. However, standard dispersive readout collapses all qubit eigenstates indiscriminately. Here, we demonstrate heralded leakage detection on a fixed-frequency transmon by applying a Rabi drive on the computational transition during dispersive readout. The drive makes the computational states indistinguishable to the probe on the resonator while leaving the second and higher excited states distinguishable from the computational states. We achieve a leakage-detection fidelity of 97.1(3)% within the 80-ns detection window, with a false-flag rate of 2.3(3)% from the computational subspace. Conditioned on the no-leakage outcome, the post-detection state retains an average fidelity of 92.9(5)% across the six cardinal states for an equal mixture of computational and leakage population. The scheme requires no circuit elements beyond those used for standard dispersive readout, making it applicable to various types of superconducting qubits without hardware modification.

quant-ph↗

arXiv2026

DRDN: Decoupled Representation Dynamic Network for From-Scratch ViT Class-Incremental Learning

Dynamic expansion methods for class-incremental learning (CIL) protect task-specific knowledge by growing dedicated tokens or subnetworks, yet our analyses suggest that classification supervision alone does not sufficiently preserve task-agnostic shared backbone representations over long incremental sequences. We identify two intertwined challenges: cross-task confusion from sequential training on predominantly current-task data, which biases decision boundaries toward recent tasks; and under-optimized shared representations in the backbone that cap long-term discriminability as tasks accumulate. We propose the Decoupled Representation Dynamic Network (DRDN), which addresses these challenges via two orthogonal mechanisms. For shared backbone representations, DRDN continuously applies masked image modeling (MIM) at every incremental step, with reconstruction gradients routed exclusively through the backbone, encouraging it to retain general visual structure beyond class-discriminative cues. For task-specific discrimination, DRDN employs hierarchical task token expansion across all transformer layers, with a modified per-task attention rule that reduces inter-task interference. We support this design with accuracy degradation analysis and cross-task confusion rate measurements. In the from-scratch ViT CIL setting (no external pretraining), DRDN consistently improves over strong token-expansion baselines with comparable backbone scale. On CIFAR100-B0 (10 steps), DRDN achieves 77.19% average accuracy, outperforming DKT by 1.36 points and DyTox by 3.53 points, with an advantage that grows at longer incremental sequences. Multi-seed validation confirms stability (+/-0.31%). The MIM decoder is active only during training, adding no inference-time parameters or computation.

cs.CV↗

arXiv2026

Retrieved Images as Visual Thought: Training-Free Multimodal In-Context Learning for the Open-vs-Closed Gap

Recent work on Thinking with Images makes vision a dynamic part of reasoning, but does so through generation: the model invokes external tools, synthesizes code, or imagines new imagery, each at the cost of a tool protocol, brittle code, or an expensive training pipeline. A fourth route makes vision dynamic without generating anything, by retrieving labeled exemplar images and reasoning over them, yet it remains underexplored despite being train-free. We present ReVisIT, a train-free framework that realizes this retrieval-based route by treating each retrieved image-label pair as a unit of visual thought. ReVisIT combines structured class definitions, per-query multimodal retrieval of exemplars, and alternating user/assistant injection of those exemplars before joint multi-attribute decoding, and degrades gracefully to whichever components a task admits. On VL-ICL Bench Fast Open MiniImageNet, Qwen3-VL-30B-A3B with ReVisIT reaches 98.5% at 4-shot, statistically indistinguishable from the 72B LLaVA-OneVision SOTA (98.7%) on this near-saturated task at about 1/2.4 the parameters, while the same backbone without the scaffold sits at chance. The turns layer alone adds 26.1 points to GPT-4.1 on free-form concept induction (Bongard-OpenWorld), and the full stack yields a 4-6 point macro gain across three backbones on MAAC-Bench, a new license-clean 27-class, 5-attribute benchmark, significant by paired bootstrap on the curator-derived attributes. Component analysis shows that retrieval-plus-turns is the universal lever while structured definitions are need-adaptive, and that 83% of the retrieval gain comes from retrieval quality rather than from the presence of exemplars. MAAC-Bench is released with a rubric-grounded LLM verification protocol that replaces author spot-check on subjective attributes.

cs.CV↗

arXiv2026

Diffusion Forcing Planner: History-Annealed Planning with Time-Dependent Guidance for Autonomous Driving

Learning-based motion planners, despite recent progress, often suffer from temporal inconsistency. Small perturbations across frames can accumulate into unstable trajectories, degrading comfort and safety in closed-loop driving. Several methods attempt to inject history as a static conditioning signal to stabilize outputs, only to induce the planner to copy historical patterns instead of adapting to environment contexts. To address this limitation, we propose Diffusion Forcing Planner (DFP), a diffusion-based planning framework driven by history-guided control. Specifically, DFP decomposes the full trajectory into history, current and future segments, and assign independent noise levels to each segment. The model jointly denoises the historical and the future segments, enforcing a heterogeneous joint diffusion process. At inference, classifier-free guidance (CFG) is applied to steer future sampling using annealed history in a controllable manner. Closed-loop evaluation and comprehensive ablations on nuPlan show that DFP achieves competitive performance while producing continuous, stable, and controllable motion plans in complex driving scenarios.

cs.RO↗

arXiv2026

Rethinking Local Learning: A Cheaper and Faster Recipe for LLM Post-Training

LLM post-training typically propagates task gradients through the full depth of the model. Although this end-to-end structure is simple and general, it couples task adaptation to full-depth activation storage, long-range backward dependencies and direct task-gradient access to pretrained representations. We argue that this full-depth backward coupling can be unnecessarily expensive and intrusive, particularly when post-training supervision is much narrower than pre-training. To this end, we propose \textbf{LoPT}: Local-Learning Post-Training, a simple post-training strategy that makes gradient reach an explicit design choice. LoPT places a single gradient boundary at the transformer midpoint: the second-half block learns from the task objective, while the first-half block is updated by a lightweight feature-reconstruction objective to preserve useful representations and maintain interface compatibility. LoPT shortens the task-induced backward path while limiting direct interference from narrow task gradients on early-layer representations. Extensive experiments demonstrate that LoPT achieves competitive performance with lower memory cost, higher training efficiency and better retention of pretrained capabilities. Our code is available at: https://github.com/HumyuShi/LoPT

cs.CL↗

arXiv2026

Correct Is Not Enough: Training Reasoning Planners with Executor-Grounded Rewards

Reinforcement learning with verifiable rewards has become a common way to improve explicit reasoning in large language models, but final-answer correctness alone does not reveal whether the reasoning trace is faithful, reliable, or useful to the model that consumes it. This outcome-only signal can reinforce traces that are right for the wrong reasons, overstate reasoning gains by rewarding shortcuts, and propagate flawed intermediate states in multi-step systems. To this end, we propose TraceLift, a planner-executor training framework that treats reasoning as a consumable intermediate artifact. During planner training, the planner emits tagged reasoning. A frozen executor turns this reasoning into the final artifact for verifier feedback, while an executor-grounded reward shapes the intermediate trace. This reward multiplies a rubric-based Reasoning Reward Model (RM) score by measured uplift on the same frozen executor, crediting traces that are both high-quality and useful. To make reasoning quality directly learnable, we introduce TRACELIFT-GROUPS, a rubric-annotated reason-only dataset built from math and code seed problems. Each example is a same-problem group containing a high-quality reference trace and multiple plausible flawed traces with localized perturbations that reduce reasoning quality or solution support while preserving task relevance. Extensive experiments on code and math benchmarks show that this executor-grounded reasoning reward improves the two-stage planner-executor system over execution-only training, suggesting that reasoning supervision should evaluate not only whether a trace looks good, but also whether it helps the model that consumes it. Our code is available at: https://github.com/MasaiahHan/TraceLift

cs.AI↗

arXiv2026

Temporal-Mode Engineering for Multiplexed Microwave Photons and Mode-Selective Quantum State Transfer

Quantum communication between distant superconducting qubits on separate chips using itinerant microwave photons has been studied to realize distributed quantum information processing. To enhance information capacity and fault tolerance in quantum networks, it is beneficial to encode a large quantity of quantum information using auxiliary degrees of freedom of these photons. In this work, we experimentally investigate the use of temporal modes of photon wave packets. Through the photon-shaping technique with a fixed-frequency transmon qubit, we generate single microwave photons in four orthogonal temporal modes propagating along a waveguide. We demonstrate mode-selective absorption across orthogonal modes via the time-reversed process of emission, achieving absorption efficiencies exceeding 0.89 for mode-matched cases, while remaining below 0.13 for orthogonal modes. Photons rejected by a given receiver mode can remain mutually orthogonal, enabling selective absorption at subsequent receivers in future multi-node architectures. These results highlight the feasibility of temporal-mode engineering for constructing a higher-dimensional orthogonal basis for multiplexed quantum networks.

quant-ph↗

arXiv2025

Deterministic Quantum Communication Between Fixed-Frequency Superconducting Qubits via Broadband Resonators

Quantum communication between remote chips is essential for realizing large-scale superconducting quantum computers. For such communication, itinerant microwave photons propagating through transmission lines offer a promising approach. However, demonstrations to date have relied on frequency-tunable circuit elements to compensate for fabrication-related parameter variations between sender and receiver devices, introducing control complexity and limiting scalability. In this work, we demonstrate deterministic quantum state transfer and remote entanglement generation between fixed-frequency superconducting qubits on separate chips. To compensate for the sender-receiver mismatch, we employ a frequency-tunable photon-generation technique which enables us to adjust the photon frequency without modifying circuit parameters. To enhance the frequency tunability, we implement broadband transfer resonators composed of two coupled coplanar-waveguide resonators, achieving a bandwidth of more than 100 MHz. This broadband design enables successful quantum communication across a 30-MHz range of photon frequencies between the remote qubits. Quantum process tomography reveals state transfer fidelities of around 79% and Bell-state fidelities of around 73% across the full frequency range. Our approach avoids the complexity of the control lines and noise channels, providing a flexible pathway toward scalable quantum networks.

quant-ph↗

arXiv2025

Generation of frequency-bin-encoded dual-rail cluster states via time-frequency multiplexing of microwave photonic qubits

Cluster states are a class of multi-qubit entangled states with broad applications such as quantum metrology and one-way quantum computing. Here, we present a protocol to generate frequency-bin-encoded dual-rail cluster states using a superconducting circuit consisting of a fixed-frequency transmon qubit, a resonator and a Purcell filter. We implement time-frequency multiplexing by sequentially emitting co-propagating microwave photons of distinct frequencies. The frequency-bin dual-rail encoding enables erasure detection based on photon occupancy. We characterize the state fidelity using quantum tomography and quantify the multipartite entanglement using the metric of localizable entanglement. Our implementation achieves a state fidelity exceeding 50$\%$ for a cluster state consisting of up to four logical qubits. The localizable entanglement remains across chains of up to seven logical qubits. After discarding the erasure errors, the fidelity exceeds 50% for states with up to eight logical qubits, and the entanglement persists across chains of up to eleven qubits. These results highlight the improved robustness of frequency-bin dual-rail encoding against photon loss compared to conventional single-rail schemes. This work provides a scalable pathway toward high-dimensional entangled state generation and photonic quantum information processing in the microwave domain.

quant-ph↗

arXiv2025

TDR: Task-Decoupled Retrieval with Fine-Grained LLM Feedback for In-Context Learning

In-context learning (ICL) has become a classic approach for enabling LLMs to handle various tasks based on a few input-output examples. The effectiveness of ICL heavily relies on the quality of these examples, and previous works which focused on enhancing example retrieval capabilities have achieved impressive performances. However, two challenges remain in retrieving high-quality examples: (1) Difficulty in distinguishing cross-task data distributions, (2) Difficulty in making the fine-grained connection between retriever output and feedback from LLMs. In this paper, we propose a novel framework called TDR. TDR decouples the ICL examples from different tasks, which enables the retrieval module to retrieve examples specific to the target task within a multi-task dataset. Furthermore, TDR models fine-grained feedback from LLMs to supervise and guide the training of the retrieval module, which helps to retrieve high-quality examples. We conducted extensive experiments on a suite of 30 NLP tasks, the results demonstrate that TDR consistently improved results across all datasets and achieves state-of-the-art performance. Meanwhile, our approach is a plug-and-play method, which can be easily combined with various LLMs to improve example retrieval abilities for ICL. The code is available at https://github.com/Nnn-s/TDR.

cs.CL↗

arXiv2025

Generation of Frequency-Tunable Shaped Single Microwave Photons Using a Fixed-Frequency Superconducting Qubit

Scaling up a superconducting quantum computer will likely require quantum communication between remote chips, which can be implemented using an itinerant microwave photon in a transmission line. To realize high-fidelity communication, it is essential to control the frequency and temporal shape of the microwave photon. In this work, we demonstrate the generation of frequency-tunable shaped microwave photons without resorting to any frequency-tunable circuit element. We develop a framework which treats a microwave resonator as a band-pass filter mediating the interaction between a superconducting qubit and the modes in the transmission line. This interpretation allows us to stimulate the photon emission by an off-resonant drive signal. We characterize how the frequency and temporal shape of the generated photon depends on the frequency and amplitude of the drive signal. By modulating the drive amplitude and frequency, we achieve a frequency tunability of 40 MHz while maintaining the photon mode shape time-symmetric. Through measurements of the quadrature amplitudes of the emitted photons, we demonstrate consistently high state and process fidelities around 95\% across the tunable frequency range. Our hardware-efficient approach eliminates the need for additional biasing lines typically required for frequency tuning, offering a simplified architecture for scalable quantum communication.

quant-ph↗

arXiv2024

A cryogenic on-chip microwave pulse generator for large-scale superconducting quantum computing

For superconducting quantum processors, microwave signals are delivered to each qubit from room-temperature electronics to the cryogenic environment through coaxial cables. Limited by the heat load of cabling and the massive cost of electronics, such an architecture is not viable for millions of qubits required for fault-tolerant quantum computing. Monolithic integration of the control electronics and the qubits provides a promising solution, which, however, requires a coherent cryogenic microwave pulse generator that is compatible with superconducting quantum circuits. Here, we report such a signal source driven by digital-like signals, generating pulsed microwave emission with well-controlled phase, intensity, and frequency directly at millikelvin temperatures. We showcase high-fidelity readout of superconducting qubits with the microwave pulse generator. The device demonstrated here has a small footprint, negligible heat load, great flexibility to operate, and is fully compatible with today's superconducting quantum circuits, thus providing an enabling technology for large-scale superconducting quantum computers.

quant-ph↗

arXiv2024

NTIRE 2024 Quality Assessment of AI-Generated Content Challenge

This paper reports on the NTIRE 2024 Quality Assessment of AI-Generated Content Challenge, which will be held in conjunction with the New Trends in Image Restoration and Enhancement Workshop (NTIRE) at CVPR 2024. This challenge is to address a major challenge in the field of image and video processing, namely, Image Quality Assessment (IQA) and Video Quality Assessment (VQA) for AI-Generated Content (AIGC). The challenge is divided into the image track and the video track. The image track uses the AIGIQA-20K, which contains 20,000 AI-Generated Images (AIGIs) generated by 15 popular generative models. The image track has a total of 318 registered participants. A total of 1,646 submissions are received in the development phase, and 221 submissions are received in the test phase. Finally, 16 participating teams submitted their models and fact sheets. The video track uses the T2VQA-DB, which contains 10,000 AI-Generated Videos (AIGVs) generated by 9 popular Text-to-Video (T2V) models. A total of 196 participants have registered in the video track. A total of 991 submissions are received in the development phase, and 185 submissions are received in the test phase. Finally, 12 participating teams submitted their models and fact sheets. Some methods have achieved better results than baseline methods, and the winning methods in both tracks have demonstrated superior prediction performance on AIGC.

cs.CV↗

arXiv2023

Photon-noise-tolerant dispersive readout of a superconducting qubit using a nonlinear Purcell filter

Residual noise photons in a readout resonator become a major source of dephasing for a superconducting qubit when the resonator is optimized for a fast, high-fidelity dispersive readout. Here, we propose and demonstrate a nonlinear Purcell filter that suppresses such an undesirable dephasing process without sacrificing the readout performance. When a readout pulse is applied, the filter automatically reduces the effective linewidth of the readout resonator, increasing the sensitivity of the qubit to the input field. The noise tolerance of the device we have fabricated is shown to be enhanced by a factor of 3 relative to a device with a linear filter. The measurement rate is enhanced by another factor of 3 by utilizing the bifurcation of the nonlinear filter. A readout fidelity of 99.4% and a quantum nondemolition fidelity of 99.2% are achieved using a 40-ns readout pulse. The nonlinear Purcell filter will be an effective tool for realizing a fast, high-fidelity readout without compromising the coherence time of the qubit.

quant-ph↗

arXiv2023

CMSG Cross-Media Semantic-Graph Feature Matching Algorithm for Autonomous Vehicle Relocalization

Relocalization is the basis of map-based localization algorithms. Camera and LiDAR map-based methods are pervasive since their robustness under different scenarios. Generally, mapping and localization using the same sensor have better accuracy since matching features between the same type of data is easier. However, due to the camera's lack of 3D information and the high cost of LiDAR, cross-media methods are developing, which combined live image data and Lidar map. Although matching features between different media is challenging, we believe cross-media is the tendency for AV relocalization since its low cost and accuracy can be comparable to the same-sensor-based methods. In this paper, we propose CMSG, a novel cross-media algorithm for AV relocalization tasks. Semantic features are utilized for better interpretation the correlation between point clouds and image features. What's more, abstracted semantic graph nodes are introduced, and a graph network architecture is integrated to better extract the similarity of semantic features. Validation experiments are conducted on the KITTI odometry dataset. Our results show that CMSG can have comparable or even better accuracy compared to current single-sensor-based methods at a speed of 25 FPS on NVIDIA 1080 Ti GPU.

cs.CV↗

arXiv2023

RCP-RF: A Comprehensive Road-car-pedestrian Risk Management Framework based on Driving Risk Potential Field

Recent years have witnessed the proliferation of traffic accidents, which led wide researches on Automated Vehicle (AV) technologies to reduce vehicle accidents, especially on risk assessment framework of AV technologies. However, existing time-based frameworks can not handle complex traffic scenarios and ignore the motion tendency influence of each moving objects on the risk distribution, leading to performance degradation. To address this problem, we novelly propose a comprehensive driving risk management framework named RCP-RF based on potential field theory under Connected and Automated Vehicles (CAV) environment, where the pedestrian risk metric are combined into a unified road-vehicle driving risk management framework. Different from existing algorithms, the motion tendency between ego and obstacle cars and the pedestrian factor are legitimately considered in the proposed framework, which can improve the performance of the driving risk model. Moreover, it requires only O(N 2) of time complexity in the proposed method. Empirical studies validate the superiority of our proposed framework against state-of-the-art methods on real-world dataset NGSIM and real AV platform.

cs.LG↗

arXiv2023

Frequency-tunable microwave quantum light source based on superconducting quantum circuits

A nonclassical light source is essential for implementing a wide range of quantum information processing protocols, including quantum computing, networking, communication, and metrology. In the microwave regime, propagating photonic qubits that transfer quantum information between multiple superconducting quantum chips serve as building blocks of large-scale quantum computers. In this context, spectral control of propagating single photons is crucial for interfacing different quantum nodes with varied frequencies and bandwidth. Here we demonstrate a microwave quantum light source based on superconducting quantum circuits that can generate propagating single photons, time-bin encoded photonic qubits and qudits. In particular, the frequency of the emitted photons can be tuned in situ as large as 200 MHz. Even though the internal quantum efficiency of the light source is sensitive to the working frequency, we show that the fidelity of the propagating photonic qubit can be well preserved with the time-bin encoding scheme. Our work thus demonstrates a versatile approach to realizing a practical quantum light source for future distributed quantum computing.

quant-ph↗

arXiv2022

An ultra-high gain single-photon transistor in the microwave regime

A photonic transistor that can switch or amplify an optical signal with a single gate photon requires strong non-linear interaction at the single-photon level. Circuit quantum electrodynamics provides great flexibility to generate such an interaction, and thus could serve as an effective platform to realize a high-performance single-photon transistor. Here we demonstrate such a photonic transistor in the microwave regime. Our device consists of two microwave cavities dispersively coupled to a superconducting qubit. A single gate photon imprints a phase shift on the qubit state through one cavity, and further shifts the resonance frequency of the other cavity. In this way, we realize a gain of the transistor up to 53.4 dB, with an extinction ratio better than 20 dB. Our device outperforms previous devices in the optical regime by several orders in terms of optical gain, which indicates a great potential for application in the field of microwave quantum photonics and quantum information processing.

quant-ph↗