Extensive Study of Multiple Deep Neural Networks for Complex Random Telegraph Signals
This work addresses the challenge of analyzing complex RTSs in physical, chemical, and biological systems, which is crucial for identifying underlying mechanisms related to performance sensitivity, but it appears incremental as it builds on existing deep learning methods for temporal data.
The authors tackled the problem of quantifying complex multilevel random telegraph signals (RTSs), which are difficult to analyze due to exponential complexity, by developing a three-step protocol using deep neural networks and demonstrated model accuracy on a large dataset with varying noise levels.
Time-fluctuating signals are ubiquitous and diverse in many physical, chemical, and biological systems, among which random telegraph signals (RTSs) refer to a series of instantaneous switching events between two discrete levels from single-particle movements. Reliable RTS analyses are crucial prerequisite to identify underlying mechanisms related to performance sensitivity. When numerous levels partake, complex patterns of multilevel RTSs occur, making their quantitative analysis exponentially difficult, hereby systematic approaches are found elusive. Here, we present a three-step analysis protocol via progressive knowledge-transfer, where the outputs of early step are passed onto a subsequent step. Especially, to quantify complex RTSs, we build three deep neural network architectures that can process temporal data well and demonstrate the model accuracy extensively with a large dataset of different RTS types affected by controlling background noise size. Our protocol offers structured schemes to quantify complex RTSs from which meaningful interpretation and inference can ensue.