Abstract:Many natural and man-made systems are prone to critical transitions -- abrupt and potentially devastating changes in dynamics. Deep learning classifiers can provide an early warning signal (EWS) for critical transitions by learning generic features of bifurcations (dynamical instabilities) from large simulated training data sets. So far, classifiers have only been trained to predict continuous-time bifurcations, ignoring rich dynamics unique to discrete-time bifurcations. Here, we train a deep learning classifier to provide an EWS for the five local discrete-time bifurcations of codimension-1. We test the classifier on simulation data from discrete-time models used in physiology, economics and ecology, as well as experimental data of spontaneously beating chick-heart aggregates that undergo a period-doubling bifurcation. The classifier outperforms commonly used EWS under a wide range of noise intensities and rates of approach to the bifurcation. It also predicts the correct bifurcation in most cases, with particularly high accuracy for the period-doubling, Neimark-Sacker and fold bifurcations. Deep learning as a tool for bifurcation prediction is still in its nascence and has the potential to transform the way we monitor systems for critical transitions.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problem of how to predict bifurcations in discrete - time systems. Specifically, the authors focus on providing early - warning signals (EWS) through deep - learning methods to identify and predict five local one - dimensional discrete - time bifurcations: period - doubling, Neimark - Sacker bifurcation, fold bifurcation, transcritical bifurcation, and pitchfork bifurcation. These problems are very important in fields such as physiology, economics, and ecology, because the dynamic changes of these systems may lead to catastrophic transitions. #### Background and Motivation Many natural and man - made systems are prone to critical transitions - that is, sudden and potentially devastating changes in dynamics. These critical transitions can occur in heart rhythms, financial markets, or ecosystems. Although traditional early - warning signals (such as variance and lag - 1 autocorrelation) can provide warnings in some cases, they cannot accurately predict specific bifurcation types and perform poorly in cases of high noise or when approaching bifurcations at a high speed. #### Research Objectives 1. **Develop new prediction tools**: Use deep - learning classifiers to predict bifurcations in discrete - time systems and provide more reliable early - warning signals. 2. **Improve prediction accuracy**: Evaluate the performance of deep - learning classifiers under different noise intensities and approaching - bifurcation rates, and compare them with traditional indicators (such as variance and lag - 1 autocorrelation). 3. **Verify the generalization ability of the model**: Test the generalization ability of the classifier through simulated data and experimental data (such as data from spontaneously beating heart cell aggregates) to ensure its effectiveness in practical applications. #### Method Overview - **Generate training data**: Use a normal - form model containing high - order terms and noise to generate a large amount of simulated data covering five discrete - time bifurcations. - **Train the classifier**: Adopt a neural network with a CNN - LSTM architecture and train two independent classifiers (Classifier 1 and Classifier 2), which learn features from different parts of the time series respectively. - **Test and evaluate**: Test the performance of the classifier on multiple theoretical models and experimental data, and evaluate its advantages over traditional indicators. Through this method, the authors hope to provide a new tool for monitoring critical transitions, thereby better preventing harmful dynamic changes and promoting favorable transitions.

Predicting discrete-time bifurcations with deep learning

Let's do the time-warp-attend: Learning topological invariants of dynamical systems

Learning from the past: predicting critical transitions with machine learning trained on surrogates of historical data

Early-warning signals for bifurcations in random dynamical systems with bounded noise

Early Warning Signals for Bifurcations Embedded in High Dimensions

Forecasting Fold Bifurcations through Physics-Informed Convolutional Neural Networks

Adapting Physics-Informed Neural Networks for Bifurcation Detection in Ecological Migration Models

Deep learning in bifurcations of particle trajectories

Prediction Of Period-Doubling Bifurcation Based On Dynamic Recognition And Its Application To Power Systems

Deep learning for predicting the occurrence of tipping points

Deep Learning for predicting rate-induced tipping

Data-Driven Bifurcation Analysis via Learning of Homeomorphism

Data-driven bifurcation analysis using parameter-dependent trajectories

Adaptable reservoir computing: A paradigm for model-free data-driven prediction of critical transitions in nonlinear dynamical systems

Precursor for Bifurcation of Axial Compression System based on Deterministic Learning

Prediction of excitable wave dynamics using machine learning

Prediction of Unobserved Bifurcation by Unsupervised Extraction of Slowly Time-Varying System Parameter Dynamics from Time Series Using Reservoir Computing

Extrapolating tipping points and simulating non-stationary dynamics of complex systems using efficient machine learning

Trans-Bifurcation Prediction of Dynamics in terms of Extreme Learning Machines with Control Inputs

Active search for Bifurcations