Graph classification aims to predict the class labels of graphs and has a wide range of applications in many real-world domains. However, most of existing graph neural networks for graph classification tasks use 90 $\%$ of labeled graphs for training and the remaining 10 $\%$ for testing, which obviously struggle in solving the problem of the scarcity of labeled graphs in real-world graph classification scenarios. And it is arduous to label a large number of graph examples for training because of the difficulty and resource consumption in the tagging process. Motivated by this, we propose a novel active and semi-supervised graph neural network (ASGNN) framework, which endeavors to complete graph classification tasks with a small number of labeled graph examples and available unlabeled graph examples. In our framework, active learning selects high-uncertain and representative graph examples from the test set and add them to the training set after annotation. Semi-supervised learning is utilized to select the high-confidence unlabeled graph examples containing structural information from the test set, and add them to the training set after pseudo labeling. To improve the generalization performance of the graph classification model, multiple GNNs are trained collaboratively for promoting the expressiveness of each other and increasing the reliability of graph classification results. Overall, the ASGNN framework takes fully use of unlabeled graph examples to reinforce graph classification effectively, and can be applied to any existing supervised graph neural networks for graph classification. Experimental results on benchmark graph datasets demonstrate that the proposed framework yields competitive performance on graph classification tasks with only a small number of labeled graph examples.
9 March 2022 in IEEE Transactions on Vehicular Technology
Radar parameter estimation in terms of its range, angle and velocity plays a crucial role in many applications. Multiple-input multiple-output (MIMO) radar with frequency diverse array (FDA) is capable of resolving targets in one directional beam with different ranges and velocities (Doppler shifts). Prevalent methods obtain target parameters in a sequential manner to get rid of exhausted multi-dimensional search, but they suffer from accumulated estimation errors. To tackle this issue, a tensor generalized weighted linear predictor (TGWLP) is devised for FDA-MIMO radar parameter estimation, where the parameters are estimated in a parallel manner. Tensor modeling of multidimensional FDA-MIMO radar signal is developed, so that the joint parameter estimation is casted into multiple-pulse-group version of three-dimensional (3D) HR problem associated to the mixed Swerling model. Pulse-group diversity is exploited to obtain precise velocity estimation. In the presence of targets with some identical parameters, the final estimations of unambiguous slant range, conic angle, and radial velocity of a moving target can be easily obtained after the parallel frequency estimations. Besides, all the parameter pairing is automatically achieved, which is free of the extra burden from pairing process. Furthermore, the identifiability of the proposed joint estimator is analyzed. Finally, theoretical analysis and simulations are included to demonstrate that the proposed approach can achieve improved performance compared to the existing methods.
22 May 2025 in ACM Transactions on Multimedia Computing, Communications, and Applications
Large pre-trained vision-language models (VLMs) like CLIP have shown great potential for solving the unsupervised domain adaptation (UDA) problem. Existing prompt learning for UDA based on the unsupervised-trained VLMs requires distribution alignment between source and target domains in the common space for both vision and language branches. However, it is difficult for rough cross-domain alignment to maintain the discriminative semantic structure of both domains. Besides, the coarse features with non-informative noises due to ignoring the pseudo-label noises may cause failures to concentrate on precise semantics alignment. In this work, we propose a Prompt-Based Invertible Mapping Alignment (PIMA) method to incorporate discriminative domain knowledge into prompt learning, which is featured with refined cross-domain alignment in two separate space with a well-kept structure. Specifically, we design an invertible neural network-based homeomorphism mapping, and then achieve distribution alignment through such invertible mapping for connecting source and target visual feature space, which can preserve the data semantic structure. For better semantic alignment in vision-language space, we develop cross-modal implicit contrastive learning module to regularize non-informative features, which aims to find the low-rankness of implicit representation space. We conducted extensive experiments on three benchmark datasets to prove the advantages of our proposed PIMA over state-of-the-art methods.
The transmit array of multi-overlapped-transmit-subarray configured bistatic multiple-input multiple-output (MOTS MIMO) radar is partitioned into a number of overlapped subarrays, which is different from the traditional bistatic MIMO radar. In this paper, a new unitary ESPRIT scheme for joint estimation of the direction of departure (DOD) and the direction of arrival (DOA) for MOTS MIMO radar is proposed. In our method, each overlapped-transmit-subarray (OTS) with the identical effective aperture is regarded as a transmit element and the characteristics that the phase delays between the two OTSs is utilized. First, the measurements corresponding to all the OTSs are partitioned into two groups which have a rotational invariance relationship with each other. Then, the properties of centro-Hermitian matrices and real-valued rotational invariance factors are exploited to double the measurement samples and reduce computational complexity. Finally, the close-formed solution of automatically paired DOAs and DODs of targets is derived in a new manner. The proposed scheme provides increased estimation accuracy with the combination of inherent advantages of MOTS MIMO radar with unitary ESPRIT. Simulation results are presented to demonstrate the effectiveness and advantage of the proposed scheme.
Spatial sparsity of the target space has been successfully exploited to provide accurate range-angle images by the methods based on sparse signal reconstruction (SSR) in Multiple-Input Multiple-Output (MIMO) radar imaging applications. The SSR based method discretizes the continuous target space into finite grid points and generates an observation model utilized in image reconstruction. However, inaccuracies in the observation model may cause various degradations and spurious peaks in the reconstructed images. In the process of the image formation, the off-grid problem frequently occurs that the true locations of targets that do not coincide with the computation grid. In this paper, we consider the case that the true location of a target has both range and angle-varying two-dimensional (2D) off-grid errors with a noninformative prior. From a variational Bayesian perspective, an iterative algorithm is developed for joint MIMO radar imaging with orthogonal frequency division multiplexing (OFDM) linear frequency modulated (LFM) waveforms and 2D off-grid error estimation of off-grid targets. The targets during multiple probing pulses are modeled as Swerling II case and a unified generalized inverse Gaussian (GIG) prior is adopted for the target reflection coefficient variance at all snapshots. Furthermore, an approach to reducing the computational workload of the signal recovery process is proposed by using singular value decomposition. Experimental results show that the proposed algorithm is insensitive to noise and has improved accuracy in terms of mean squared estimation error under different computation grid interval.