Predictive audio representations for early detection and tracking of hidden dynamic objects
研究通过自监督预训练和多任务微调的方法，利用音频信号预测隐藏动态物体的数量、类型及方向，以解决自动驾驶中被遮挡交通参与者难以及时检测的问题。
0 citationsRead paper
Selected work
Representative Papers
CalibBEV: LiDAR-Camera Calibration via BEV Alignment
This work addresses the challenging problem of cross-modal calibration between LiDAR and cameras by proposing the first bird’s-eye-view (BEV)-based alignment framework. The method unifies multimodal data into a shared BEV space and employs a two-stage optimization strategy: it first implicitly regresses coarse calibration parameters and then explicitly aligns cross-modal features, enhanced by a CLIP-inspired contrastive loss to enforce semantic consistency. By integrating domain-specific BEV feature extraction with contrastive learning constraints, the approach significantly outperforms existing methods, achieving state-of-the-art calibration accuracy. On the KITTI and nuScenes benchmarks, it reduces relative rotation error by 51% and 68%, and translation error by 80% and 91%, respectively.
0 citationsRead paper
Approximate Quantum State Preparation Through Proximal Policy Optimization
This work addresses the challenge of quantum state preparation, where the exponentially growing search space with increasing qubit count renders the discovery of optimal quantum circuits intractable. To overcome this, the authors propose a deep reinforcement learning framework based on Proximal Policy Optimization (PPO), wherein an agent incrementally constructs quantum circuits to approximate a target state while minimizing gate count. This approach introduces deep reinforcement learning into quantum architecture search and achieves, for the first time, joint optimization of high-fidelity state preparation and circuit compactness. Experimental results demonstrate that the method attains approximation fidelities as high as $10^{-14}$ across various predefined and randomly generated target states in systems ranging from 2 to 5 qubits.
0 citationsRead paper
Explaining Weather Bulletins via ILP
This work addresses the challenge of enhancing the interpretability of weather forecasts by automatically explaining the rationale behind meteorologists’ selection of specific graphical symbols in weather bulletins. Leveraging the FastLAS2 framework, the study extracts Answer Set Programming (ASP) facts from synthetic meteorological data and real bulletins issued by OSMER FVG to construct inductive logic programming (ILP) examples, from which concise and human-readable logical rules are learned. It pioneers the application of non-monotonic ILP to the task of meteorological bulletin interpretation, enabling a symbolic and generalizable model of expert decision-making. The proposed approach not only accurately uncovers the underlying logic governing symbol selection but also demonstrates robust generalizability across different geographical regions and data sources.
0 citationsRead paper
Parallel QEC Decoding Applied to Distributed Quantum Computing
This work addresses the challenges of high computational complexity and limited scalability in quantum error correction (QEC) decoding for distributed quantum computing by proposing a parallel decoding approach that integrates belief propagation with ordered statistics decoding. The method innovatively incorporates singular value decomposition into the belief propagation framework to perform localized preprocessing of error vectors within subregions of the lattice, thereby enabling efficient parallelization. By doing so, it achieves substantial reductions in computational complexity while maintaining high decoding accuracy, significantly enhancing scalability. This approach offers a practical and efficient QEC decoding solution well-suited for large-scale distributed quantum computing architectures.
0 citationsRead paper
Recent publications
Latest Papers
Predictive audio representations for early detection and tracking of hidden dynamic objects
研究通过自监督预训练和多任务微调的方法，利用音频信号预测隐藏动态物体的数量、类型及方向，以解决自动驾驶中被遮挡交通参与者难以及时检测的问题。
0 citationsRead paper
CalibBEV: LiDAR-Camera Calibration via BEV Alignment
This work addresses the challenging problem of cross-modal calibration between LiDAR and cameras by proposing the first bird’s-eye-view (BEV)-based alignment framework. The method unifies multimodal data into a shared BEV space and employs a two-stage optimization strategy: it first implicitly regresses coarse calibration parameters and then explicitly aligns cross-modal features, enhanced by a CLIP-inspired contrastive loss to enforce semantic consistency. By integrating domain-specific BEV feature extraction with contrastive learning constraints, the approach significantly outperforms existing methods, achieving state-of-the-art calibration accuracy. On the KITTI and nuScenes benchmarks, it reduces relative rotation error by 51% and 68%, and translation error by 80% and 91%, respectively.
0 citationsRead paper
Approximate Quantum State Preparation Through Proximal Policy Optimization
This work addresses the challenge of quantum state preparation, where the exponentially growing search space with increasing qubit count renders the discovery of optimal quantum circuits intractable. To overcome this, the authors propose a deep reinforcement learning framework based on Proximal Policy Optimization (PPO), wherein an agent incrementally constructs quantum circuits to approximate a target state while minimizing gate count. This approach introduces deep reinforcement learning into quantum architecture search and achieves, for the first time, joint optimization of high-fidelity state preparation and circuit compactness. Experimental results demonstrate that the method attains approximation fidelities as high as $10^{-14}$ across various predefined and randomly generated target states in systems ranging from 2 to 5 qubits.
0 citationsRead paper
Explaining Weather Bulletins via ILP
This work addresses the challenge of enhancing the interpretability of weather forecasts by automatically explaining the rationale behind meteorologists’ selection of specific graphical symbols in weather bulletins. Leveraging the FastLAS2 framework, the study extracts Answer Set Programming (ASP) facts from synthetic meteorological data and real bulletins issued by OSMER FVG to construct inductive logic programming (ILP) examples, from which concise and human-readable logical rules are learned. It pioneers the application of non-monotonic ILP to the task of meteorological bulletin interpretation, enabling a symbolic and generalizable model of expert decision-making. The proposed approach not only accurately uncovers the underlying logic governing symbol selection but also demonstrates robust generalizability across different geographical regions and data sources.
0 citationsRead paper
Parallel QEC Decoding Applied to Distributed Quantum Computing
This work addresses the challenges of high computational complexity and limited scalability in quantum error correction (QEC) decoding for distributed quantum computing by proposing a parallel decoding approach that integrates belief propagation with ordered statistics decoding. The method innovatively incorporates singular value decomposition into the belief propagation framework to perform localized preprocessing of error vectors within subregions of the lattice, thereby enabling efficient parallelization. By doing so, it achieves substantial reductions in computational complexity while maintaining high decoding accuracy, significantly enhancing scalability. This approach offers a practical and efficient QEC decoding solution well-suited for large-scale distributed quantum computing architectures.
0 citationsRead paper