Machine-learning-based method for fiber-bending eavesdropping detection.

In this Letter, we present a scheme for detecting fiber-bending eavesdropping based on feature extraction and machine learning (ML). First, 5-dimensional features from the time-domain signal are extracted from the optical signal, and then a long short-term memory (LSTM) network is applied for eavesdropping and normal event classification. Experimental data are collected from a 60 km single-mode fiber transmission link with eavesdropping implemented by a clip-on coupler.

Effects of nine typical technologies for primary autonomous vehicles on road safety in China.

The autonomous vehicle is profoundly changing the future of transportation safety. The reduction in collisions with different injury degrees and the savings of crash-related economic costs if nine autonomous vehicle technologies were promoted to be wildly available in China are evaluated. The quantitative analysis was divided into three main parts: (1) Calculate the technical effectiveness of nine autonomous vehicle technologies in collisions through a systematic literature review; (2) Apply the technical effectiveness to estimate the potential effects on avoiding collisions and saving crash-related economic costs in China if all vehicles had these technologies; and (3) Quantify the influence of current technical limitations in speed applicability, weather applicability, light applicability, and active rate on potential effects.

Security analysis of a QAM modulated quantum noise stream cipher under a correlation attack.

Quantum noise stream cipher where encrypted signals are masked by quantum noise and ASE noise provides a physical layer of security. It requires the transmitter and the receiver to share a stream cipher that is generated from a PRNG. Yet a correlation attack threatens its security due to the mathematical properties of PRNG.

Exploring Weakly Labeled Images for Video Object Segmentation With Submodular Proposal Selection.

Video object segmentation (VOS) is important for various computer vision problems, and handling it with minimal human supervision is highly desired for the large-scale applications. To bring down the supervision, existing approaches largely follow a data mining perspective by assuming the availability of multiple videos sharing the same object categories. It, however, would be problematic for the tasks that consume a single video.