In this paper, we propose a first-order discrete model for random early detection (RED) with nonlinearity, focusing on the average queue size in a router. The RED is a representative router-based algorithm for congestion control. Congestion is avoided by discarding packets at random using the packet drop probability function and the average queue size in RED. Our proposed model replaces the linear packet drop probability function of the original RED with a nonlinear function and can continuously change the nonlinearity strength, namely the bending degree of the nonlinear function. In other words, the model mathematically analyzes the influence of the nonlinearity strength in the packet drop probability function on the control of RED. Our simulation results showed the reproducibility of the model under three traffic conditions: light, heavy, and extremely heavy, by comparing the model and the network simulator-2. Furthermore, we performed a fixed point analysis on the steady-states to determine the average queue size. Consequently, the effect of the change in the nonlinearity strength on the average queue size was mathematically ensured.

In this study, we verify the robustness of using parameter space estimation with ridge regression to predict a critical transition. The parameter space can be estimated from only two time-series data sets generated by a system with different parameter values. Thereby, we can predict the parameter value at which the critical transition will occur by plotting a bifurcation diagram in the estimated parameter space. We are able to show that this method can predict the critical transition from time-series data sets perturbed by several noise intensities. In numerical experiments, we verify the robustness for several noise intensities while adjusting a normalization parameter of the ridge regression. Additionally, we confirm the differences in the trained synaptic weights between when the predictions are successful and when we are unable to consistently obtain a successful prediction.

Some studies started analyzing time series via complex networks in the noughties. A study reported that time series, which were transformed into networks via recurrence plots (RPs), especially k-nearest neighbor RP, exhibited superfamily phenomena resulting from the network motif analysis. However, it is not hard to imagine that the transformation properties drastically change the results of the motif analysis. In this study, we examine that by introducing another type of RP.As a result, the RP employed in this study works better for the superfamily phenomena among the time series than the previously used RP.

This paper proposes a novel lifting-based lossless image coding using CNN predictors and context estimators optimized by JADE. Our method adopts five types of CNN predictors for enabling not only CNN prediction based on the nature of the input image but also an optimal prediction for efficient compression can be realized. Introducing new optimizable parameters for context estimation enables further improvement of context estimation. Also, the efficiency of arithmetic coding is enhanced by introducing a grouping algorithm considering predictor utilization. The encoding experiments on various images support that the proposed method outperforms well-known existing lossless coding methods.

In broadband communication systems, an analog-to-digital converter (ADC) with few-bit quantization is a promising component in a RF front-end to ensure higher sampling rate. To mitigate the nonlinear behavior in the few-bit ADC, a stochastic resonance (SR), which improves a symbol error rate (SER) in the presence of the noise, has been focused. The SER degrades in a high signal-to-noise ratio (SNR) region. In this paper, we introduce the addition of the intentional noise in the SR-based receiver. To add the intentional noise appropriately, we propose simple estimation methods for the channel noise-power in the receiver with the few-bit ADC. To obtain the closed-form for a error function, we employ a sigmoid approximation. Numerical examples show that the proposed receiver can estimate the noise-power in a specific range of SNR. Low SER can be achieved for higher SNR by the addition of the intentional noise.