Mingyu Ma, Jiun-Yu Kao, Arpit Gupta, Yu-Hsiang Lin, Wenbo Zhao, Tagyoung Chung, Wei Wang, Kai-Wei Chang, Nanyun Peng. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.
15 December 2022 in IEEE Signal Processing Letters
In this letter, we predict the locations as probability distributions for the tasks of image object detection. We adopt the Kullback-Leibler divergences as the regression losses to train the deep neural networks. Since most existing evaluations label the objects with rectangular bounding boxes, we propose the Nearest Distribution Converter to find the closest uniform distributions from the predicted ones. Our proposed method can improve the detected accuracy measured in mAP by 0.57%, 0.75%, and 0.48% on the models YOLOv3, the YOLOv4-tiny, and the YOLOv4, respectively.
Small cells play an important bridging role for low-latency and high-speed communication in early 5G era. The short-term availability forecasting in small cell networks based on ensemble neural networks was proposed to improve the wireless network quality and customer experience. In this research, the standard specifications of Broadband Forum data model and 3GPP LTE key performance indicator were the input factor in ensemble neural networks to predict the availability in small cell networks. In addition, the network logs were also collected as the input factor of the forecasting model. In experimental result, the accuracy of validation dataset was more than 90%, and the device’s KPI changes were closely related to the malfunction. The device’s parameters could be adjusted and optimized in advance according to the forecasting result to avoid malfunction occurs.
Models of various NLP tasks have been shown to exhibit stereotypes, and the bias in the question answering (QA) models is especially harmful as the output answers might be directly consumed by the end users. There have been datasets to evaluate bias in QA models, while bias mitigation technique for the QA models is still under-explored. In this work, we propose BMBI, an approach to mitigate the bias of multiple-choice QA models. Based on the intuition that a model would lean to be more biased if it learns from a biased example, we measure the bias level of a query instance by observing its influence on another instance. If the influenced instance is more biased, we derive that the query instance is biased. We then use the bias level detected as an optimization objective to form a multi-task learning setting in addition to the original QA task. We further introduce a new bias evaluation metric to quantify bias in a comprehensive and sensitive way. We show that our method could be applied to multiple QA formulations across multiple bias categories. It can significantly reduce the bias level in all 9 bias categories in the BBQ dataset while maintaining comparable QA accuracy.