Qin Shi

Jiawei Du, Qin Shi, Joey Tianyi Zhou

Dataset distillation is a newly emerging task that synthesizes a small-size dataset used in training deep neural networks (DNNs) for reducing data storage and model training costs. The synthetic datasets are expected to capture the essence of the knowledge contained in real-world datasets such that the former yields a similar performance as the latter. Recent advancements in distillation methods have produced notable improvements in generating synthetic datasets. However, current state-of-the-art methods treat the entire synthetic dataset as a unified entity and optimize each synthetic instance equally. This static optimization approach may lead to performance degradation in dataset distillation. Specifically, we argue that static optimization can give rise to a coupling issue within the synthetic data, particularly when a larger amount of synthetic data is being optimized. This coupling issue, in turn, leads to the failure of the distilled dataset to extract the high-level features learned by the deep neural network (DNN) in the latter epochs. In this study, we propose a new dataset distillation strategy called Sequential Subset Matching (SeqMatch), which tackles this problem by adaptively optimizing the synthetic data to encourage sequential acquisition of knowledge during dataset distillation. Our analysis indicates that SeqMatch effectively addresses the coupling issue by sequentially generating the synthetic instances, thereby enhancing its performance significantly. Our proposed SeqMatch outperforms state-of-the-art methods in various datasets, including SVNH, CIFAR-10, CIFAR-100, and Tiny ImageNet. Our code is available at https://github.com/shqii1j/seqmatch.

Qin Shi, Amber Yijia Zheng, Qifan Song et al.

We propose the task of knowledge distillation detection, which aims to determine whether a student model has been distilled from a given teacher, under a practical setting where only the student's weights and the teacher's API are available. This problem is motivated by growing concerns about model provenance and unauthorized replication through distillation. To address this task, we introduce a model-agnostic framework that combines data-free input synthesis and statistical score computation for detecting distillation. Our approach is applicable to both classification and generative models. Experiments on diverse architectures for image classification and text-to-image generation show that our method improves detection accuracy over the strongest baselines by 59.6% on CIFAR-10, 71.2% on ImageNet, and 20.0% for text-to-image generation. The code is available at https://github.com/shqii1j/distillation_detection.

Qin Shi, Xiaowei Cui, Sihao Zhao et al.

Localizing moving targets in unknown harsh environments has always been a severe challenge. This letter investigates a novel localization system based on multi-agent networks, where multiple agents serve as mobile anchors broadcasting their time-space information to the targets. We study how the moving target can localize itself using the sequential time of arrival (TOA) of the one-way broadcast signals. An extended two-step weighted least squares (TSWLS) method is proposed to jointly estimate the position and velocity of the target in the presence of agent information uncertainties. We also address the large target clock offset (LTCO) problem for numerical stability. Analytical results reveal that our method reaches the Cramer-Rao lower bound (CRLB) under small noises. Numerical results show that the proposed method performs better than the existing algorithms.

Qin Shi, Xiaowei Cui, Sihao Zhao et al.

High-accuracy absolute localization for a team of vehicles is essential when accomplishing various kinds of tasks. As a promising approach, collaborative localization fuses the individual motion measurements and the inter-vehicle measurements to collaboratively estimate the states. In this paper, we focus on the range-only collaborative localization, which specifies the inter-vehicle measurements as inter-vehicle ranging measurements. We first investigate the observability properties of the system and derive that to achieve bounded localization errors, two vehicles are required to remain static like external infrastructures. Under the guide of the observability analysis, we then propose our range-only collaborative localization system which categorize the ground vehicles into two static vehicles and dynamic vehicles. The vehicles are connected utilizing a UWB network that is capable of both producing inter-vehicle ranging measurements and communication. Simulation results validate the observability analysis and demonstrate that collaborative localization is capable of achieving higher accuracy when utilizing the inter-vehicle measurements. Extensive experimental results are performed for a team of 3 and 5 vehicles. The real-world results illustrate that our proposed system enables accurate and real-time estimation of all vehicles' absolute poses.

Qin Shi, Xiaowei Cui, Wei Li et al.

Navigation applications relying on the Global Navigation Satellite System (GNSS) are limited in indoor environments and GNSS-denied outdoor terrains such as dense urban or forests. In this paper, we present a novel accurate, robust and low-cost GNSS-independent navigation system, which is composed of a monocular camera and Ultra-wideband (UWB) transceivers. Visual techniques have gained excellent results when computing the incremental motion of the sensor, and UWB methods have proved to provide promising localization accuracy due to the high time resolution of the UWB ranging signals. However, the monocular visual techniques with scale ambiguity are not suitable for applications requiring metric results, and UWB methods assume that the positions of the UWB transceiver anchor are pre-calibrated and known, thus precluding their application in unknown and challenging environments. To this end, we advocate leveraging the monocular camera and UWB to create a map of visual features and UWB anchors. We propose a visual-UWB Simultaneous Localization and Mapping (SLAM) algorithm which tightly combines visual and UWB measurements to form a joint non-linear optimization problem on Lie-Manifold. The 6 Degrees of Freedom (DoF) state of the vehicles and the map are estimated by minimizing the UWB ranging errors and landmark reprojection errors. Our navigation system starts with an exploratory task which performs the real-time visual-UWB SLAM to obtain the global map, then the navigation task by reusing this global map. The tasks can be performed by different vehicles in terms of equipped sensors and payload capability in a heterogeneous team. We validate our system on the public datasets, achieving typical centimeter accuracy and 0.1% scale error.

Qin Shi, Xiaowei Cui, Sihao Zhao et al.

The spatiotemporal information plays crucial roles in a multi-agent system (MAS). However, for a highly dynamic and dense MAS in unknown environments, estimating its spatiotemporal states is a difficult problem. In this paper, we present BLAS: a wireless broadcast relative localization and clock synchronization system to address these challenges. Our BLAS system exploits a broadcast architecture, under which a MAS is categorized into parent agents that broadcast wireless packets and child agents that are passive receivers, to reduce the number of required packets among agents for relative localization and clock synchronization. We first propose an asynchronous broadcasting and passively receiving (ABPR) protocol. The protocol schedules the broadcast of parent agents using a distributed time division multiple access (D-TDMA) scheme and delivers inter-agent information used for joint relative localization and clock synchronization. We then present distributed state estimation approaches in parent and child agents that utilize the broadcast inter-agent information for joint estimation of spatiotemporal states. The simulations and real-world experiments based on ultra-wideband (UWB) illustrate that our proposed BLAS cannot only enable accurate, high-frequency and real-time estimation of relative position and clock parameters but also support theoretically an unlimited number of agents.