Searcharxiv⌕ Search

arXiv subjects

Xinran Chen

Publications and source records attributed to Xinran Chen.

20 records · Page 2Linked to original sources

Improving Non-autoregressive Machine Translation with Error Exposure and Consistency Regularization

Being one of the IR-NAT (Iterative-refinemennt-based NAT) frameworks, the Conditional Masked Language Model (CMLM) adopts the mask-predict paradigm to re-predict the masked low-confidence tokens. However, CMLM suffers from the data distribution discrepancy between training and inference, where the observed tokens are generated differently in the two cases. In this paper, we address this problem with the training approaches of error exposure and consistency regularization (EECR). We construct the mixed sequences based on model prediction during training, and propose to optimize over the masked tokens under imperfect observation conditions. We also design a consistency learning method to constrain the data distribution for the masked tokens under different observing situations to narrow down the gap between training and inference. The experiments on five translation benchmarks obtains an average improvement of 0.68 and 0.40 BLEU scores compared to the base models, respectively, and our CMLMC-EECR achieves the best performance with a comparable translation quality with the Transformer. The experiments results demonstrate the effectiveness of our method.

cs.CL↗

Predictive and Recommendatory Spectrum Decision for Cognitive Radio

Cognitive radio technology enables improving the utilization efficiency of the precious and scarce radio spectrum. How to maximize the overall spectrum efficiency while minimizing the conflicts with primary users is vital to cognitive radio. The key is to make the right decisions of accessing the spectrum. Spectrum prediction can be employed to predict the future states of a spectrum band using previous states of the spectrum band, whereas spectrum recommendation recommends secondary users a subset of available spectrum bands based on secondary user's previous experiences of accessing the available spectrum bands. In this paper, a framework for spectrum decision based on spectrum prediction and spectrum recommendation is proposed. As a benchmark, a method based on extreme learning machine (ELM) for single-user spectrum prediction and a method based on Q-learning for multiple-user spectrum prediction are proposed. At the stage of spectrum decision, two methods based on Q-learning andMarkov decision process (MDP), respectively, are also proposed to enhance the overall performance of spectrum decision. Experimental results show that the performance of the spectrum decision framework is much better.

cs.NI↗