Infrared small target images are not conducive to detection due to their low visibility. Although there are many algorithms have been proposed to solve this problem, existing methods inevitably struggle to strike a good balance between efficiency and performance. This letter proposes a fast infrared small target detection algorithm with a low computational time consumption, only 0.012 s, while maintaining excellent detection performance. There are only two steps in the proposed method. First, the original image is processed by dilate operation with a ring structuring element. Then, the final detection result is calculated by global contrast measure (GCM) between the original image and the processed image according to the ratio form and the difference form. The experimental results demonstrate that the proposed method consistently achieves a robust and effective performance on six real sequences. In particular, the speed of the proposed method is at least two times faster than the state-of-the-art methods.

Transformers are powerful for sequence modeling. Nearly all state-of-the-art language models and pre-trained language models are based on the Transformer architecture. However, it distinguishes sequential tokens only with the token position index. We hypothesize that better contextual representations can be generated from the Transformer with richer positional information. To verify this, we propose a segment-aware Transformer (Segatron), by replacing the original token position encoding with a combined position encoding of paragraph, sentence, and token. We first introduce the segment-aware mechanism to Transformer-XL, which is a popular Transformer-based language model with memory extension and relative position encoding. We find that our method can further improve the Transformer-XL base model and large model, achieving 17.1 perplexity on the WikiText-103 dataset. We further investigate the pre-training masked language modeling task with Segatron. Experimental results show that BERT pre-trained with Segatron (SegaBERT) can outperform BERT with vanilla Transformer on various NLP tasks, and outperforms RoBERTa on zero-shot sentence representation learning. Our code is available on GitHub.

Incorporating review information into the recommender system has been demonstrated to be an effective method for boosting the recommendation performance. Previous research mainly focus on designing advanced architectures to better profile the users and items. However, the review information in realities can be highly sparse and imbalanced, which poses great challenges for effective user/item representations and satisfied performance enhancement. To alleviate this problem, in this paper, we propose to improve review-based recommendation by counterfactually augmenting the training samples. We focus on a common setting --- feature-aware recommendation, and the main building block of our idea lies in the counterfactual question: "what would be the user's decision if her feature-level preference had been different?''. When augmenting the training samples, we actively change the user preference (also called intervention), and predict the user feedback on the items based on pre-trained recommender models. Instead of changing the user preference in a random manner, we design a learning-based method to discover the samples which are more effective for model optimization. In order to improve the sample qualities, we propose two strategies --- constrained feature perturbation and frequency-based sampling --- to equip our model. Since the sample generation model can be not perfect, we theoretically analyze the relation between the model prediction error and the number of generated samples. As a byproduct, our framework can explain the user pair-wise preference, which is complementary to the traditional point-wise explanations. Extensive experiments demonstrate that our model can significantly improve the performance of the state-of-the-art methods.

Mohan Zhang, Luchen Tan, Zihang Fu, Kun Xiong, Jimmy Lin, Ming Li, Zhengkai Tu. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021.

This article presents a novel transformable wheeled robot with three motion modes. Based on the four-bar mechanism design, the transformable wheel can transform into CW (clockwise) legged wheel mode and CCW (counterclock-wise) legged wheel mode from the circular wheel. Both legged wheel modes achieve good obstacle-climbing performance when the robot overcomes obstacles such as steps in the forward and backward directions. Mathematical expression of shape of the transformable wheel is established for design variables selection of the wheel to improve the obstacle-overcoming ability of the legged wheel modes. Dynamics model of the robot in the step-overcoming scene is analyzed to determine and reduce the torque load of the motor. Finally, two prototypes of the robot are developed. The early-stage prototype made by 3D printing verifies the step-overcoming ability of the transformable wheel. Then, structural strength of the robot chassis and the robustness of the hardware are improved in the second prototype, which can play a role in SAR and indoor service tasks. In general, the prototype can overcome steps with the height of more than 2 times the radius of the wheel in CW/CCW legged wheel mode and move at a maximum speed of 3.15 m/s in circular wheel mode (about 5.83 times the length of the robot per second).

Image captioning (IC) is a commonly-used technique for generating textual image description, which finds its applications on semantic image retrieval and multi-modal image understanding, among many others. This paper focuses on an important IC method specialized for generating aesthetic descriptions of images, i.e., aesthetic image captioning (AIC). Despite some effectiveness of initial work on AIC, their performances are inherently limited due to a lack of consideration of user preferences on aesthetics and better aesthetic feature, making it unusable for real-world applications where human users present a large variation on evaluating visual aesthetics of images. To tackle this, we propose a novel personalized aesthetic image caption (PAIC) approach for capturing and incorporating user preferences for AIC tasks. Our approach mainly contains Aesthetic feature Extraction Network(AEN), User Encoder network(UEN) and a personalized image caption model. AEN is designed to extract more expressive feature, UEN is introduced for learning the user vector from the limited information in our AVA-PCap dataset. Personalized image caption model is constructed to generate the caption when given the user id and photo pairs. The experimental results show that our methods outperform baselines by 10% , which is encouraging for a first step towards personalized aesthetic image caption.

We tackle the problem of question answering directly on a large document collection, combining simple “bag of words” passage retrieval with a BERT-based reader for extracting answer spans. In the context of this architecture, we present a data augmentation technique using distant supervision to automatically annotate paragraphs as either positive or negative examples to supplement existing training data, which are then used together to fine-tune BERT. We explore a number of details that are critical to achieving high accuracy in this setup: the proper sequencing of different datasets during fine-tuning, the balance between “difficult” vs. “easy” examples, and different approaches to gathering negative examples. Experimental results show that, with the appropriate settings, we can achieve large gains in effectiveness on two English and two Chinese QA datasets. We are able to achieve results at or near the state of the art without any modeling advances, which once again affirms the cliché “there’s no data like more data”.

Understanding user preference is of key importance for an effective recommender system. For comprehensive user profiling, many efforts have been devoted to extract user feature-level preference from the review information. Despite effectiveness, existing methods mostly assume linear relationships among the users, items, and features, and the collaborative information is usually utilized in an implicit and insufficient manner, which limits the recommender capacity in modeling users’ diverse preferences. For bridging this gap, in this article, we propose to formulate user feature-level preferences by a neural signed hypergraph and carefully design the information propagation paths for diffusing collaborative filtering signals in a more effective manner. By taking the advantages of the neural model’s powerful expressiveness, the complex relationship patterns among users, items, and features are sufficiently discovered and well utilized. By infusing graph structure information into the embedding process, the collaborative information is harnessed in a more explicit and effective way. We conduct comprehensive experiments on real-world datasets to demonstrate the superiorities of our model.

The next generation network, 5G, is expected to provide service-oriented networks where different applications are served in isolation according to their own requirements. It is challenging to have an efficient common resource allocation mechanism for virtual networks (VNs) that have different objectives. In this work, we propose a dynamic reservation and deep reinforcement learning based autonomous virtual resource management. The infrastructure provider periodically reserves the unused resource to the VNs based on their ratio of minimum resource requirements. Then, the VNs autonomously control their resource allocation by using deep reinforcement learning based on the average quality of service utility and resource utilization of their users. With the defined design pattern in this paper, virtual operators can customize their own utility function and objective function based on their own requirements. We simulate our work to show resource utilization and satisfaction of the VNs.