arXiv subjects

Zikun Xu

Publications and source records attributed to Zikun Xu.

8 recordsLinked to original sources

arXiv2026

Reparametrizing Shampoo and SOAP for Subspace Basis Updates and BFloat16 Storage

Shampoo-based methods, such as KL-Shampoo and SOAP, have demonstrated strong performance in training neural networks and rely on QR decomposition. Because existing QR implementations require single-precision (FP32) arithmetic and remain computationally expensive, these methods become time- and memory-intensive when their preconditioning matrices are large. Moreover, using BFloat16 (BFP16) storage to reduce memory usage can degrade the performance of Shampoo-based methods. We propose a reparametrization of the preconditioner that supports BFP16 storage and forms a complete basis by combining updated basis vectors with unchanged ones. By updating only part of the basis through QR decomposition in a subspace, our approach reduces computational overhead while mitigating the performance degradation caused by BFP16 storage. Our approach applies broadly to Shampoo-based methods that employ QR decomposition, including KL-Shampoo, SOAP, and KL-SOAP. In particular, it improves the performance of SOAP and KL-SOAP under BFP16 storage, enabling KL-SOAP to match or exceed KL-Shampoo. Overall, our approach makes Shampoo-based methods more memory- and time-efficient.

cs.LG↗

arXiv2026

Stable Fine-Time-Step Long-Horizon Turbulence Prediction with a Multi-Stepsize Mixture-of-Experts Neural Operator

Neural operators have been increasingly used as data-driven surrogates for time-marching predictions of turbulent flows. However, long-horizon autoregressive prediction is sensitive to error accumulation and the choice of prediction interval. Excessively small time increments may increase temporal redundancy and lengthen rollouts, which can degrade the stability of neural operators in turbulence forecasting. This work pursues a unified objective: stable long-horizon autoregressive prediction at fine temporal resolution for three-dimensional turbulence. We propose a multi-stepsize mixture-of-experts (Ms-MoE) neural operator built on an implicit factorized Transformer (IFactFormer) backbone. The model conditions on a requested relative stride and uses a time-step router to activate scale-specific routed experts together with a shared expert, yielding a single architecture that represents a family of stride-parameterized time-advancement operators. We evaluate the approach on forced homogeneous isotropic turbulence (HIT) and turbulent channel flow using filtered direct numerical simulation datasets. Relative to sampling intervals used in previous studies, we construct training datasets with up to 20 times finer temporal resolution and report long-horizon autoregressive rollouts using qualitative time-slice comparisons and long-time-averaged statistics. Ms-MoE-IFactFormer yields more stable long-horizon rollouts and improved agreement with long-time-averaged statistics on both HIT and turbulent channel flow, suggesting potential for stable time-marching at fine temporal resolution in more complex turbulent flows.

physics.flu-dyn↗

arXiv2026

Long-SCOPE: Fully Sparse Long-Range Cooperative 3D Perception

Cooperative 3D perception via Vehicle-to-Everything communication is a promising paradigm for enhancing autonomous driving, offering extended sensing horizons and occlusion resolution. However, the practical deployment of existing methods is hindered at long distances by two critical bottlenecks: the quadratic computational scaling of dense BEV representations and the fragility of feature association mechanisms under significant observation and alignment errors. To overcome these limitations, we introduce Long-SCOPE, a fully sparse framework designed for robust long-distance cooperative 3D perception. Our method features two novel components: a Geometry-guided Query Generation module to accurately detect small, distant objects, and a learnable Context-Aware Association module that robustly matches cooperative queries despite severe positional noise. Experiments on the V2X-Seq and Griffin datasets validate that Long-SCOPE achieves state-of-the-art performance, particularly in challenging 100-150 m long-range settings, while maintaining highly competitive computation and communication costs.

cs.CV↗

arXiv2025

RaFD: Flow-Guided Radar Detection for Robust Autonomous Driving

Radar has shown strong potential for robust perception in autonomous driving; however, raw radar images are frequently degraded by noise and "ghost" artifacts, making object detection based solely on semantic features highly challenging. To address this limitation, we introduce RaFD, a radar-based object detection framework that estimates inter-frame bird's-eye-view (BEV) flow and leverages the resulting geometric cues to enhance detection accuracy. Specifically, we design a supervised flow estimation auxiliary task that is jointly trained with the detection network. The estimated flow is further utilized to guide feature propagation from the previous frame to the current one. Our flow-guided, radar-only detector achieves achieves state-of-the-art performance on the RADIATE dataset, underscoring the importance of incorporating geometric information to effectively interpret radar signals, which are inherently ambiguous in semantics.

cs.RO↗

arXiv2025

Understanding and Improving Shampoo and SOAP via Kullback-Leibler Minimization

Shampoo and its efficient variant, SOAP, employ structured second-moment estimations and have shown strong performance for training neural networks (NNs). In practice, however, Shampoo typically requires step-size grafting with Adam to be competitive, and SOAP mitigates this by applying Adam in Shampoo's eigenbasis -- at the cost of additional memory overhead from Adam in both methods. Prior analyses have largely relied on the Frobenius norm to motivate these estimation schemes. We instead recast their estimation procedures as covariance estimation under Kullback-Leibler (KL) divergence minimization, revealing a previously overlooked theoretical limitation and motivating principled redesigns. Building on this perspective, we develop $\textbf{KL-Shampoo}$ and $\textbf{KL-SOAP}$, practical schemes that match or exceed the performance of Shampoo and SOAP in NN pre-training while achieving SOAP-level per-iteration runtime. Notably, KL-Shampoo does not rely on Adam to attain competitive performance, eliminating the memory overhead introduced by Adam. Across our experiments, KL-Shampoo consistently outperforms SOAP, Shampoo, and even KL-SOAP, establishing the KL-based approach as a promising foundation for designing structured methods in NN optimization. An implementation of KL-Shampoo/KL-SOAP is available at https://github.com/yorkerlin/KL-Methods

stat.ML↗

arXiv2024

MergeOcc: Bridge the Domain Gap between Different LiDARs for Robust Occupancy Prediction

LiDAR-based 3D occupancy prediction evolved rapidly alongside the emergence of large datasets. Nevertheless, the potential of existing diverse datasets remains underutilized as they kick in individually. Models trained on a specific dataset often suffer considerable performance degradation when deployed to real-world scenarios or datasets involving disparate LiDARs. This paper aims to develop a generalized model called MergeOcc, to simultaneously handle different LiDARs by leveraging multiple datasets. The gaps among LiDAR datasets primarily manifest in geometric disparities and semantic inconsistencies. Thus, MergeOcc incorporates a novel model featuring a geometric realignment module and a semantic label mapping module to enable multiple datasets training (MDT). The effectiveness of MergeOcc is validated through experiments on two prominent datasets for autonomous vehicles: OpenOccupancy-nuScenes and SemanticKITTI. The results demonstrate its enhanced robustness and remarkable performance across both types of LiDARs, outperforming several SOTA multi-modality methods. Notably, despite using an identical model architecture and hyper-parameter set, MergeOcc can significantly surpass the baseline due to its exposure to more diverse data. MergeOcc is considered the first cross-dataset 3D occupancy prediction pipeline that effectively bridges the domain gap for seamless deployment across heterogeneous platforms.

cs.CV↗

arXiv2024

4D Millimeter-Wave Radar in Autonomous Driving: A Survey

The 4D millimeter-wave (mmWave) radar, proficient in measuring the range, azimuth, elevation, and velocity of targets, has attracted considerable interest within the autonomous driving community. This is attributed to its robustness in extreme environments and the velocity and elevation measurement capabilities. However, despite the rapid advancement in research related to its sensing theory and application, there is a conspicuous absence of comprehensive surveys on the subject of 4D mmWave radar. In an effort to bridge this gap and stimulate future research, this paper presents an exhaustive survey on the utilization of 4D mmWave radar in autonomous driving. Initially, the paper provides reviews on the theoretical background and progress of 4D mmWave radars, encompassing aspects such as the signal processing workflow, resolution improvement approaches, and extrinsic calibration process. Learning-based radar data quality improvement methods are present following. Then, this paper introduces relevant datasets and application algorithms in autonomous driving perception, localization and mapping tasks. Finally, this paper concludes by forecasting future trends in the realm of 4D mmWave radar in autonomous driving. To the best of our knowledge, this is the first survey specifically dedicated to the 4D mmWave radar in autonomous driving.

eess.SP↗

arXiv2022

Quantifying relation between mobility patterns and socioeconomic status of dockless sharing-bike users

Bikes are among the healthiest, greenest, and most affordable means of transportation for a better future city, but mobility patterns of riders with different income were rarely studied due to limitations on collecting data. Newly emergent dockless bike-sharing platforms that record detailed information regarding each trip provide us a unique opportunity. Attribute to its better usage flexibility and accessibility, dockless bike-sharing platforms are booming over the past a few years worldwide and reviving the riding fashion in cities. In this work, by exploiting massive riding records in two megacities from a dockless bike-sharing platform, we reveal that individual mobility patterns, including radius of gyration and average travel distance, are similar among users with different income, which indicates that human beings all follow similar physical rules. However, collective mobility patterns, including average range and diversity of visitation, and commuting directions, all exhibit different behaviors and spatial patterns across income categories. Hotspot locations that attract more cycling activities are quite different over groups, and locations where users reside are of a low user ratio for both higher and lower income groups. Lower income groups are inclined to visit less flourishing locations, and commute towards the direction to the city center in both cities, and of a smaller mobility diversity in Beijing but a larger diversity in Shanghai. In addition, differences on mobility patterns among socioeconomic categories are more evident in Beijing than in Shanghai. Our findings would be helpful on designing better promotion strategies for dockless bike-sharing platforms and towards the transition to a more sustainable green transportation.

physics.soc-ph↗