Search

Text adversarial attack is an effective strategy to investigate the vulnerability of Natural Language Processing (NLP) models. Most of text attack studies focus on word-level attacks with static or dynamic optimization algorithms. However, they are hard to balance (1) attack performance (i.e., attack success rate, word substitution rate) and (2) attack efficiency. Generally, static optimization is fast but suffers from low attack performance, and the dynamic adversary improves the attacking quality but is time-consuming. To address these challenges, a Generation-based Parallel Particle Swarm Optimization (GP 2 SO) is proposed for the adversarial text attack. Specifically, the GP 2 SO employs an adaptive strategy to determine the word modification priority, which produces a high attack performance owing to the aggressive objective function. To achieve time efficiency, we parallelize the PSO on multiple pipelines in a generation-overlapping manner. Extensive experiments on four public text recognition datasets are conducted by attacking four deep models to evaluate the effectiveness of the GP 2 SO. Experimental results manifest that the proposed GP 2 SO averagely improves the time efficiency by 272% with only 0.3% success rate reduction compared to the PSO. Besides, the GP 2 SO also shows superiorities in adversarial training and transferability compared with baselines. The code is provided to ensure reproducibility https://github.com/OutdoorManofML/GPPSO. [ABSTRACT FROM AUTHOR]

• We conduct the first attempt to the data-to-text generation in Social Internet of Things (SIoT). • We propose a novel model to automatically generate the natural language sequence description from the nonlinguistic input data in SIoT. • The proposed model integrates the traditional pipeline modules with the end-to-end neural network model. • The experimental results demonstrate that our model outperforms several other existing solutions. Social Internet of Things (SIoT) is an emerging paradigm connecting the Internet of things and social networks, enabling the social functions expanded into human-to-thing or inter-thing. There is an urgent issue of how to realize the social contact for communications between human and things or between things. This requires the natural language generation (NLG), focusing on generating text in human languages from non-linguistic data, which is of vital importance to achieve the interaction without barriers between human and other smart objects in SIoT. However, the existing solutions for data-to-text transformation rely on the specific template and standard neural networks models, which are not flexible when applying to the cross-domain data with diverse meaning. In this paper, we propose a novel model, called pipeline assisted neural networks (PAN), which integrates the traditional pipeline modules and neural generation systems, to conduct data-to-text generation tasks in SIoT. Our new PAN model makes use of attention module and gate mechanism to complete content selection, by analyzing records correlation and filtering redundant records. Then, a dedicated neural network is developed for salient span selection, pointing and transition to achieve content planning. In the end, we conduct experiments on the ROTOWIRE dataset to evaluate the performance of PAN. The results demonstrate that our proposed PAN model outperforms the existing solutions. [ABSTRACT FROM AUTHOR]