Incorporating Uncertainty into a Regression Neural Network Enables Identification of Decadal State-Dependent Predictability

Gordon, Emily M; Barnes, Elizabeth A.

doi:10.1002/essoar.10510836.2

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this work, we choose to assess both types of uncertainty and use distributions for the output layer, as well as for the network parameters of the hidden layers. Our BNN approach therefore provides a more holistic view than previous work to assess uncertainty in large‐scale ocean neural network predictions in Gordon and Barnes (2022) where a deterministic neural network is used to predict the mean and variance of the output distribution.…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Explainable Artificial Intelligence for Bayesian Neural Networks: Toward Trustworthy Predictions of Ocean Dynamics

Clare

Sonnewald

Lguensat

et al. 2022

J Adv Model Earth Syst

View full text Add to dashboard Cite

The trustworthiness of neural networks is often challenged because they lack the ability to express uncertainty and explain their skill. This can be problematic given the increasing use of neural networks in high stakes decision-making such as in climate change applications. We address both issues by successfully implementing a Bayesian Neural Network (BNN), where parameters are distributions rather than deterministic, and applying novel implementations of explainable AI (XAI) techniques. The uncertainty analysis from the BNN provides a comprehensive overview of the prediction more suited to practitioners' needs than predictions from a classical neural network. Using a BNN means we can calculate the entropy (i.e. uncertainty) of the predictions and determine if the probability of an outcome is statistically significant. To enhance trustworthiness, we also spatially apply the two XAI techniques of Layer-wise Relevance Propagation (LRP) and SHapley Additive exPlanation (SHAP) values. These XAI methods reveal the extent to which the BNN is suitable and/or trustworthy. Using two techniques gives a more holistic view of BNN skill and its uncertainty, as LRP considers neural network parameters, whereas SHAP considers changes to outputs. We verify these techniques using comparison with intuition from physical theory. The differences in explanation identify potential areas where new physical theory guided studies are needed.

show abstract

Section: Methodsmentioning

confidence: 99%

“…where N l is the number of possible variable outcomes and p ij is the probability of each outcome j for sample i (Goodfellow et al, 2016). Hence, the larger the entropy value, the less skewed the distribution and the more uncertain the model is of the result.…”

Section: Bayesian Neural Networkmentioning

confidence: 99%

Explainable Artificial Intelligence for Bayesian Neural Networks: Toward Trustworthy Predictions of Ocean Dynamics

Clare

Sonnewald

Lguensat

et al. 2022

J Adv Model Earth Syst

View full text Add to dashboard Cite

show abstract

A range of outcomes: the combined effects of internal variability and anthropogenic forcing on regional climate trends over Europe

Deser

Phillips

2023

Nonlin. Processes Geophys.

View full text Add to dashboard Cite

Abstract. Disentangling the effects of internal variability and anthropogenic forcing on regional climate trends remains a key challenge with far-reaching implications. Due to its largely unpredictable nature on timescales longer than a decade, internal climate variability limits the accuracy of climate model projections, introduces challenges in attributing past climate changes, and complicates climate model evaluation. Here, we highlight recent advances in climate modeling and physical understanding that have led to novel insights about these key issues. In particular, we synthesize new findings from large-ensemble simulations with Earth system models, observational large ensembles, and dynamical adjustment methodologies, with a focus on European climate.

show abstract

Incorporating Uncertainty into a Regression Neural Network Enables Identification of Decadal State-Dependent Predictability

Cited by 2 publications

References 0 publications

Explainable Artificial Intelligence for Bayesian Neural Networks: Toward Trustworthy Predictions of Ocean Dynamics

Explainable Artificial Intelligence for Bayesian Neural Networks: Toward Trustworthy Predictions of Ocean Dynamics

A range of outcomes: the combined effects of internal variability and anthropogenic forcing on regional climate trends over Europe

Contact Info

Product

Resources

About