Owing to recent advances in artificial intelligence and internet of things (IoT) technologies, collected big data facilitates high computational performance, while its computational resources and energy cost are large. Moreover, data are often collected but not used. To solve these problems, we propose a framework for a computational model that follows a natural computational system, such as the human brain, and does not rely heavily on electronic computers. In particular, we propose a methodology based on the concept of `computation harvesting', which uses IoT data collected from rich sensors and leaves most of the computational processes to real-world phenomena as collected data. This aspect assumes that large-scale computations can be fast and resilient. Herein, we perform prediction tasks using real-world road traffic data to show the feasibility of computation harvesting. First, we show that the substantial computation in traffic flow is resilient against sensor failure and real-time traffic changes due to several combinations of harvesting from spatiotemporal dynamics to synthesize specific patterns. Next, we show the practicality of this method as a real-time prediction because of its low computational cost. Finally, we show that, compared to conventional methods, our method requires lower resources while providing a comparable performance.
Reservoir computing derived from recurrent neural networks is more applicable to real world systems than deep learning because of its low computational cost and potential for physical implementation. Specifically, physical reservoir computing, which replaces the dynamics of reservoir units with physical phenomena, has recently received considerable attention. In this study, we propose a method of exploiting the dynamics of road traffic as a reservoir, and numerically confirm its feasibility by applying several prediction tasks based on a simple mathematical model of the traffic flow.
We investigate prediction accuracy for time series of Echo state networks with respect to several kinds of activation functions. As a result, we found that some kinds of activation functions with an appropriate nonlinearity show high performance compared to the conventional sigmoid function.
In this paper, we address the problem of reconstructing a time-domain signal (or a phase spectrogram) solely from a magnitude spectrogram. Since magnitude spectrograms do not contain phase information, we must restore or infer phase information to reconstruct a time-domain signal. One widely used approach for dealing with the signal reconstruction problem was proposed by Griffin and Lim. This method usually requires many iterations for the signal reconstruction process and depending on the inputs, it does not always produce high-quality audio signals. To overcome these shortcomings, we apply a learning-based approach to the signal reconstruction problem by modeling the signal reconstruction process using a deep neural network and training it using the idea of a generative adversarial network. Experimental evaluations revealed that our method was able to reconstruct signals faster with higher quality than the Griffin-Lim method.