Ren-Song Tsay

Wei-Yi Kuo, Ren-Song Tsay

As data collection and analysis become critical functions for many cloud applications, proper data sharing with approved parties is required. However, the traditional data sharing scheme through centralized data escrow servers may sacrifice owners' privacy and is weak in security. Mainly, the servers physically own all data while the original data owners have only virtual ownership and lose actual access control. Therefore, we propose a 3-layer SSE-ABE-AES (3LSAA) cryptography-based privacy-protected data-sharing protocol based on the assumption that servers are honest-but-curious. The 3LSAA protocol realizes automatic access control management and convenient file search even if the server is not trustable. Besides achieving data self-sovereignty, our approach also improves system usability, eliminates the defects in the traditional SSE and ABE approaches, and provides a local AES key recovery method for user's availability.

Zheng-Xun Jiang, Ren-Song Tsay

In this paper we propose a double-linked blockchain data structure that greatly improves blockchain performance and guarantees single chain with no forks. Additionally, with the proposed proof-of-refundable-tax (PoRT) consensus algorithm, our approach can construct highly reliable, efficient, fair and stable blockchain operations. The PoRT algorithm adopts a verifiable random function instead of mining to select future block maintainers with the probability proportional to each participant's personal refundable tax. The individual refundable tax serves as an index of the activeness of participation and hence PoRT can effectively prevent Sybil attacks. Also, with the block-completion reward deducted from each maintainer's refundable tax, our blockchain system maintains a stable wealth distribution and avoids the "rich become richer" problem. We have implemented the approach and tested with very promising results.

Hsuan-Yi Lin, Ren-Song Tsay

In this paper, we present a systematic approach that transforms the program execution trace into the frequency domain and precisely identifies program phases. The analyzed results can be embedded into program code to mark the starting point and execution characteristics, such as CPI (Cycles per Instruction), of each phase. The so generated information can be applied to runtime program phase prediction. With the precise program phase information, more intelligent software and system optimization techniques can be further explored and developed.

Hsu-Hsun Chin, Ren-Song Tsay, Hsin-I Wu

Recent convolutional neural network (CNN) development continues to advance the state-of-the-art model accuracy for various applications. However, the enhanced accuracy comes at the cost of substantial memory bandwidth and storage requirements and demanding computational resources. Although in the past the quantization methods have effectively reduced the deployment cost for edge devices, it suffers from significant information loss when processing the biased activations of contemporary CNNs. In this paper, we hence introduce an adaptive high-performance quantization method to resolve the issue of biased activation by dynamically adjusting the scaling and shifting factors based on the task loss. Our proposed method has been extensively evaluated on image classification models (ResNet-18/34/50, MobileNet-V2, EfficientNet-B0) with ImageNet dataset, object detection model (YOLO-V4) with COCO dataset, and language models with PTB dataset. The results show that our 4-bit integer (INT4) quantization models achieve better accuracy than the state-of-the-art 4-bit models, and in some cases, even surpass the golden full-precision models. The final designs have been successfully deployed onto extremely resource-constrained edge devices for many practical applications.

Tsung-Ying Lu, Hsu-Hsun Chin, Hsin-I Wu et al.

In this paper, we propose a very compact embedded CNN processor design based on a modified logarithmic computing method using very low bit-width representation. Our high-quality CNN processor can easily fit into edge devices. For Yolov2, our processing circuit takes only 0.15 mm2 using TSMC 40 nm cell library. The key idea is to constrain the activation and weight values of all layers uniformly to be within the range [-1, 1] and produce low bit-width logarithmic representation. With the uniform representations, we devise a unified, reusable CNN computing kernel and significantly reduce computing resources. The proposed approach has been extensively evaluated on many popular image classification CNN models (AlexNet, VGG16, and ResNet-18/34) and object detection models (Yolov2). The hardware-implemented results show that our design consumes only minimal computing and storage resources, yet attains very high accuracy. The design is thoroughly verified on FPGAs, and the SoC integration is underway with promising results. With extremely efficient resource and energy usage, our design is excellent for edge computing purposes.