Convolutional Neural Networks (CNN) is widely applied in modern machine learning and pattern recognition area. Streaming hardware accelerator is an efficient design that reduces power dissipation by lowering off-chip memory access. Streaming architecture requires putting the entire network on-chip, which is a challenging task for resource-constrained embedded systems. This paper presents a streaming hardware accelerator to implement SkyNet, a lightweight neural network to address real-time object detection. The net is retrained and quantized to a very low bit in order to put all the weights into the on-chip BRAMs. We designed a highly balanced pipeline based on each layer’s calculation to improve resource utilization. We demonstrate the streaming-based accelerator on the Ultra96v2 board and achieve a throughput of 536 frames per second (FPS) at 333MHz and intersection over Union (IoU) of 73.1% on the DAC-SDC dataset. The overall computational efficiency of our accelerator is 98%. Compared to the state-of-the-art, our work achieves 2.5× throughput, 2.7× energy efficiency.

Yaojie Lu, Hongyu Lin, Jin Xu, Xianpei Han, Jialong Tang, Annan Li, Le Sun, Meng Liao, Shaoyi Chen. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Labeling a text document is usually time consuming because it requires the annotator to read the whole document and check its relevance with each possible class label. It thus becomes rather expensive to train an effective model for text classification when it involves a large dataset of long documents. In this paper, we propose an active learning approach for text classification with lower annotation cost. Instead of scanning all the examples in the unlabeled data pool to select the best one for query, the proposed method automatically generates the most informative examples based on the classification model, and thus can be applied to tasks with large scale or even infinite unlabeled data. Furthermore, we propose to approximate the generated example with a few summary words by sparse reconstruction, which allows the annotators to easily assign the class label by reading a few words rather than the long document. Experiments on different datasets demonstrate that the proposed approach can effectively improve the classification performance while significantly reduce the annotation cost.