Teaching machines to anticipate catastrophes

Lapeyrolerie, Marcus; Boettiger, Carl

doi:10.1073/pnas.2115605118

Cited by 6 publications

(8 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…cusp and Bogdanov-Takens), and bifurcations of attractors. For systems on attractors that explore a large portion of their phase space, empirical dynamical modelling 41 , reservoir computing 42 , 43 and deep neural networks 44 can be used to make forecasts that may help predict critical transitions. In cases where spatial information is available, concepts from statistical physics may be useful 45 , particularly in combination with deep learning 27 .…”

Section: Discussionmentioning

confidence: 99%

Predicting discrete-time bifurcations with deep learning

Bury,

Dylewsky,

Bauch

et al. 2023

Nat Commun

View full text Add to dashboard Cite

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 for critical transitions by learning generic features of bifurcations 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 early warning signal for the five local discrete-time bifurcations of codimension-one. 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 shows higher sensitivity and specificity than commonly used early warning signals 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.

show abstract

Section: Discussionmentioning

confidence: 99%

Predicting discrete-time bifurcations with deep learning

Bury,

Dylewsky,

Bauch

et al. 2023

Nat Commun

View full text Add to dashboard Cite

show abstract

“…Given the difficulty of forecasting never‐before‐observed behaviour, as illustrated by the Hopf and saddle‐node bifurcation scenarios, there is good reason for research to focus more on the kind of qualitative predictions long emphasized in the literature on early warning signals and resilience (Scheffer et al, 2012). Recently, ML techniques developed for classification rather than the ML methods used in regression and forecasting models considered here have demonstrated a more nuanced ability to reliably detect different classes of critical transitions in time‐series data (Bury et al, 2021; Lapeyrolerie & Boettiger, 2021). Rather than seeking to provide managers with quantitative, probabilistic forecasts reflecting a broad uncertainty in possible outcomes, this literature has sought to emphasize only a more qualitative form of prediction, such as establishing whether a system is either ‘resilient’ or ‘approaching a critical transition’.…”

Section: Discussionmentioning

confidence: 99%

“…We have created a github repository https://github.com/boettiger-lab/mee_tipping_point_forecasting that contains the code used to produce the figures herein. This repository has been archived at https://zenodo.org/badge/latestdoi/470212861 (Lapeyrolerie & Boettiger, 2022).…”

Section: Peer Reviewmentioning

confidence: 99%

“…Recent work such as Bury et al (2021) has introduced ML methods to consider classification of this transition in four possible categories (Hopf, saddle‐node, transcritical, or no bifurcation) rather than two (bifurcation or not). These are important results with considerable promise (Lapeyrolerie & Boettiger, 2021), but which nevertheless address a very different question using very different methods. Early warning signals only predict ‘a big change may be coming soon’—they do not try to forecast when or how big.…”

Section: Introductionmentioning

confidence: 96%

See 1 more Smart Citation

Limits to ecological forecasting: Estimating uncertainty for critical transitions with deep learning

Lapeyrolerie

Boettiger

2022

Methods Ecol Evol

Self Cite

View full text Add to dashboard Cite

In the current age of a rapidly changing environment, it is becoming increasingly important to study critical transitions and how to best anticipate them. Critical transitions pose extremely challenging forecasting problems, which necessitate informative uncertainty estimation rather than point forecasts. In this study, we apply some of the most cutting edge deep learning methods for probabilistic time series forecasting to several classic ecological models that examine critical transitions. Our analysis focuses on three different simulated examples of critical transitions: a Hopf bifurcation, a saddle‐node bifurcation and a stochastic transition. For each scenario, we compare the forecasts from four deep learning models, long‐short term memory networks, gated recurrent unit networks, lock recurrent neural networks and transformers, to forecasts from an ARIMA model and a MCMC estimated model that is given the true transition dynamics. We found that the deep learning models were able to perform comparably to the idealized MCMC model on the stochastic transition case, and generally in between the MCMC and ARIMA models on the Hopf and saddle‐node bifurcation examples. Our results establish that deep learning methods warrant further exploration on the challenging class of critical transition forecasting problems.

show abstract

“… 2 ), while for the deep learning method developed by Bury et al ( 1 ) a particular type of detrending is necessary because all training examples were detrended using it. Bury et al ( 1 ) and Lapeyrolerie and Boettiger ( 3 ) note that the training set would have to be expanded substantially to include richer dynamical behavior than fold, transcritical, and Hopf bifurcations. With this note, we suggest that other aspects of the training, including preprocessing, also need careful consideration.…”

mentioning

confidence: 99%