The prediction of parking space availability plays a crucial role in information systems providing parking guidance. However, controversy persists regarding the efficiency and accuracy of mainstream time series prediction methods, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). In this study, a comparison was made between a temporal convolutional network (TCN) based on CNNs and a long short-term memory (LSTM) network based on RNNs to determine an appropriate baseline for predicting parking space availability. Subsequently, a multi-head attention (MAT) mechanism was incorporated into an LSTM network, attempting to improve its accuracy. Experiments were conducted on three real and two synthetic datasets. The results indicated that the TCN achieved the fastest convergence, whereas the MAT-LSTM method provided the highest average accuracy, namely 0.0330 and 1.102 × 10⁻⁶, on the real and synthetic datasets, respectively. Furthermore, the improved MAT-LSTM model accomplished an increase of up to 48% in accuracy compared with the classic LSTM model. Consequently, we concluded that RNN-based networks are better suited for predicting long-time series. In particular, the MAT-LSTM method proposed in this study holds higher application value for predicting parking space availability with a higher accuracy.

Partial multi-label learning (PML) aims to train a classifier on dataset whose instances are over-annotated with not only relevant labels but also irrelevant labels, which is common when datasets are collected from crowd-sourcing platform. Existing works primarily approach it from a curriculum learning perspective, leveraging the memorization effect to disambiguate noisy labels and produce robust predictions. However, these methods are based on non-adaptive weighting functions and lack theoretical guidance for optimal weighting. To overcome these issues, a calibrated disambiguation model named PML-CD is proposed. We firstly formulate the optimal weighting function for curriculum-based disambiguation, which is equivalent to the calibration of the model's predicted confidences, thus provide a guidance for curriculum designing. To obtain the optimal weighting function from PML dataset during the training, a transferable calibrator is designed, which takes the histogram of positive samples' confidences as input, and outputs the optimal curriculum weighting for training. Prototype alignment regularization is also proposed to promote the model's performance. Experiments conducted on Pascal VOC, MS-COCO, NUS-WIDE and CUB have verified that our method outperforms existing state-of-the-art PML methods.

Miniature film-structured robots achieved remarkable locomotion performance and various applications. The integration of a jumping function could empower existing film-structured terrestrial robots to conquer complex terrains. However, current repeated jumping mechanisms for miniature soft robots are often constrained to specific materials and mechanical configurations, rendering them incompatible with film-structured robots. A solution that effectively addresses these limitations remains absent in the field. This work introduces an electrically actuated solution for repeated jumping in miniature soft robots, utilizing a bistable film structure coupled with a numerical two-rod model. The bistable film structure integrates a flexible film, polydimethylsiloxane (PDMS) linear spring, and a shape memory alloy spring to achieve repeated jumping motion through snap-through buckling. The numerical model is employed to optimize the spring parameters, tailoring the structure to the specific properties of the target robot. To demonstrate the practicality of this solution, the bistable film structure is seamlessly integrated into an existing fast-steering insect robot, which previously lacks jumping capabilities. By customizing PDMS spring design based on the robot body’s dimensions and stiffness, the prototype robot, measuring 23 × 16 × 8 mm³ in size and 120 mg in weight, demonstrates repeated jumping ability with a maximum horizontal jumping distance of 11.8 cm (5.1 body lengths), a jumping height of 3.8 cm (4.7 body heights) and a jumping frequency of 0.1 Hz. The robot traverses a 43 cm-long road with half walls and trench obstacles in 41 s. This work presents the bistable film structure’s potential as a repeated jumping solution for film-structured robots, enhancing their obstacle-crossing abilities and expanding their applicability in complex environments.

With the continuous development of large cross-modal models, this approach to addressing cross-modal retrieval challenges has become mainstream in the field of information retrieval. However, two major issues still persist in cross-modal retrieval: the intermodal semantic gap and the generalization of cross-modal models. The first issue is particularly challenging. Although multi-modal large models can help obtain unified semantic representations of different modalities in a shared embedding space, they still struggle to fully capture image representations, which limits their ability to understand natural language. In this work, we propose a Cross-modal Retrieval Based on a Multi-modal Large Model with Convolutional Attention and Adversarial Training (LMMCa-AT). The algorithm enhances the model's ability to learn high-dimensional semantics of images through the interaction between convolutional neural networks (CNN) and Transformers, while further reducing the number of model parameters. First, to enable images to fully comprehend text semantics, we introduce a convolutional attention module, which consists of CNN and Transformer, by improving the image encoder of the large cross-modal model CLIP. Furthermore, to enhance the robustness of the cross-modal retrieval model, we also propose an adversarial training module (ATM), which uses an adversarial perturbation generation mechanism for each modality to perform adversarial training in the embedding space, learning invariant semantics specific to both the image and text, to improve model generalization. Compared with cross-modal retrieval methods based on CNN or Transformer, the LMMCa-AT achieves an average improvement of 8.4% in image retrieval R@1 and 5.6% in text retrieval R@1 on the Ms-CoCo 1K dataset. On the Flickr 1K dataset, image retrieval R@1 improved by an average of 3.94%, and text retrieval R@1 improved by an average of 7.1%. Experimental results show the effectiveness of the convolutional attention module in understanding images, leading to improved cross-modal retrieval accuracy.