Search

With the integration of edge computing frameworks and federated learning protocols, an increasing number of copyright protection methods for deep learning models have been proposed. However, solely verifying ownership from the sender's perspective does not provide assistance to the receiver. Numerous studies have indicated that in client-edge-cloud federated learning systems, malicious users attempt to gain access to pubic models without contributing or even poison the pubic models. Therefore, it is necessary to provide a model ownership verification scheme for the receiver. Building upon existing neural network watermarking schemes, this paper proposes a dual-verification model watermarking scheme based on certification files. It introduces a certification file generation step and implements dual ownership verification of the model through adjustments in the model structure. The feasibility, robustness, and improvement in watermark embedding rate of the scheme are verified through experiments. [ABSTRACT FROM AUTHOR]

Most sinograms (i.e., Chinese characters) are phonograms (phonetic compounds). A phonogram is composed of a semantic radical and a phonetic radical, with the former usually implying the meaning of the phonogram, and the latter providing cues to its pronunciation. This study focused on the sub-lexical processing of semantic radicals which are themselves free standing sinograms. Two primed naming experiments were carried out to examine whether the meanings and pronunciations of the semantic radicals embedded in phonograms were activated or not during sinogram recognition. In Experiment 1, semantically opaque phonograms were used as primes. We observed facilitatory priming effects for targets which were semantically related to the semantic radicals embedded in primes, but not to the primes themselves. These effects were present for low-frequency primes, but not for high-frequency primes. Experiment 2 used only low-frequency phonograms as primes. We observed facilitatory priming effects for targets which were homophones of the semantic radicals embedded in primes, but not of the primes themselves. These results suggest that sub-lexical semantic and phonological information of semantic radicals are activated, and that the activation processes are modulated by the lexical frequency of the host phonograms. Our study shows that sub-lexical processing of semantic radicals is similar to that of phonetic radicals, indicating no fundamental difference between sub-lexical processing of semantic and phonetic radicals, supporting the view that a radical has a unique representation irrespective of its function in the orthographic system of Taft's model.

Purpose: This study investigates the impact of intertalker variations on the process of mapping acoustic variations on tone categories in two different tone languages. Method: Pitch stimuli manipulated from four voice ranges were presented in isolation through a blocked-talker design. Listeners were instructed to identify the stimuli that they heard as lexical tones in their native language. Results: Tone identification of Mandarin listeners exhibited relatively stable normalization regardless of the voice, whereas tone identification of Cantonese listeners was unstable and susceptible to the influence of intertalker variations. In the case of Cantonese listeners, intertalker variations had a larger effect on the perception of F0 height dimension than of F0 slope dimension. Conclusion: The comparison between Cantonese and Mandarin listeners' performances reveals an interaction of intertalker variations and the types of tone contrasts in each language. For Cantonese tones, which depend heavily on F0 height distinctions, intertalker variations result in F0 overlapping and, consequently, ambiguities among them in isolated tone perception. For Mandarin tones, which are distinctive in terms of their F0 contours, the differences in F0 contours alone seem sufficient to elicit reliable tone identification. Intertalker variations therefore have relatively limited effect on Mandarin tone perception.

To meet the watermarking requirement in encrypted domain, this paper proposed a novel reversible watermarking algorithm in histogram difference domain for encrypted image by utilizing a combination of histogram shifting and homomorphic encryption. It could embeded a watermark into the encrypted image directly to realize the process of plain image-encrypted image-watermark embedding, which simplified the traditional redundant process of plain image-encrypted image-decryption--watermark embedding-encryption again. Simulation experimental results demonstrate that the proposed algorithm can provide greater payload capacity and higher image fidelity. After it extracts the watermark completely, it can recover the original image reversibly. It outperforms other related algorithms. [ABSTRACT FROM AUTHOR]

Objective: To explore the clinical value of Lens culinaris agglutinin-reactive alpha-fetoprotein in the differentiation diagnosis between benign and malignant liver diseases, as well as the early warning of hepatocellular carcinoma., Methods: Alpha-fetoprotein variants from 300 patients with liver diseases were isolated with micro-spin column equipped lens culinaris agglutinin (LCA). The AFP and AFP-L3 were detected by the electrochemical luminescence (ECL) method, and the proportions of AFP-L3 were calculated., Results: The positive rates of AFP-L3 of HCC patients and chronic liver disease patients were 95% and 64% respectively, there were significant difference in two groups (chi2 = 134.72, P < 0.01), the HCC incidence rates of AFP-L3 positive and negative chronic liver disease patients showed significant difference (chi2 = 80.158, P < 0.01). there were no correlations between the proportion of AFP-L3 and AFP consistency(r = 0.046, P > 0.05)., Conclusions: The detection of AFP-L3 by micro-spin column assay show great clinical value in the differentiation diagnosis of benign and malignant liver diseases, as well as the early warning of hepatocellular carcinoma.