Publication

Abstract : Study of critical transitions and early warning measures are of great importance for dealing with any complex system. Manually selected statistical features with handpicked parameters have been used in a wide variety of fields for this purpose. We envision the use of deep learning architectures like simple feed forward networks (FFN), convolutional neural networks (CNN) and long short-term memory networks (LSTM) to predict these critical transitions from raw time-series data obtained from complex systems with minimal human interference in parameter choosing. As a first step towards this goal, in this study we use the above mentioned deep learning architectures to classify the states of a modified Van der Pol oscillator. We observe that the deep learning architectures produce good classification results and show promise as a tool for detection of critical transitions from raw time-series data.