Deep learning has achieved great success in various applications with the help of a large scale of datasets. As a result, sharing the valuable big data that can be applied to training deep learning models is of essential importance currently. However, how to claim ownership and protect the copyright of image big data during the sharing process is still a vital issue that should be addressed. The application of digital watermarks can protect the copyright of image data, at the same time, it also degrades the image quality at the same time. As for invisible digital watermarks, the higher the watermark embedding intensity causes the greater the host image changes. Therefore, the performance of deep learning models may decline due to using the watermarked training set. In this paper, we evaluate the influences of different embedding intensities of various watermarking algorithms on several mainstream models and conclude how the watermarking intensity affects the model training. Besides, referring to watermarking algorithms that have been proposed, we proposed a novel discrete Fourier transform-based watermarking algorithm that can achieve image copyright protection yet maintain the utility of models.
Subsurface karst systems represent substantial but underexplored methane sinks, yet the identities and activities of cave-dwelling methanotrophs remain poorly characterized. We detected increased methane oxidation rates from 2.9 ± 0.1 to 90.7 ± 4.5 ng·g −1 ·hour −1 while supplied with 2 to 500 parts per million (ppm) CH 4 to cave sediments. Atmospheric methanotroph Upland Soil Clusters γ (USCγ), responsible for this oxidation, was further assigned to three genera within the family Candidatus ( Ca .) Methyloligotrophaceae, including two previously unrecognized genera. Nano-scale secondary ion mass spectrometry (NanoSIMS) imaging and the produced 13 C-PLFAs (phospholipid fatty acids) and 13 CO 2 in 13 CH 4 -fed microcosm confirmed methane as both carbon and energy sources. These methanotrophs exhibited low half-saturation constant ( Km ; 138.8 ± 15.8 ppm), high carbon assimilation efficiency (>50%), and metabolic versatility, as revealed by metagenomics and metatranscriptomics analyses. By extrapolating global distribution of Ca. Methyloligotrophaceae and comparing methane oxidation rates between caves and soil ecosystems, we conservatively estimate that subsurface karst in southwest China sequester ~0.56 Tg CH 4 annually. These findings highlight the ecological importance of karst ecosystems as a previously overlooked methane sink.
The widespread and mature application of deep learning technology on human behavior analysis leads to the possibility that the Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) may not be able to accurately determine whether or not the user is human, which makes the distributed denial-of-service (DDoS) attacks launched by deep learning-based malicious automatic computer programs become possible. The CAPTCHA generated by adversarial examples can defend against the automatic identification by the deep learning-based method, however, if the deployed CAPTCHAs are all adversarial examples solely, the adversary has the chance to collect sufficient adversarial examples, which also causes security issues. In this paper, a user behavior-based random distribution scheme for adversarial example generated CAPTCHA is proposed to tackle the above sole distribution problem. Specifically, we generated two kinds of CAPTCHAs, one is normal and the other is generated by the fast gradient signed method (FGSM), i.e., the adversarial examples. Meanwhile, user behaviors are analyzed and associated with a reasonable probability model. According to the probability selection, the normal CAPTCHAs or strong CAPTCHAs will be delivered to the users. The experimental results illustrate that our scheme has an excellent ability to distinguish computers from humans. Thus, it can protect the CAPTCHA systems from DDoS attacks.