Uncertainty in Deep Learning, p.2 (Methods)
Table 1: List of popular uncertainty methods for deep learning models. The leftmost column lists the uncertainty methods, categorized into sampling-based and sampling-free approaches. Each column represents different aspects of these methods:
(1) Training time relative to training a single model (“Train. T.”),
(2) Inference time relative to a single model (“Inf. T.”),
(3) Memory requirements for the method (“Mem.”),
(4) A subjective metric indicating the ease of integrating the method into existing architectures and training pipelines (“Integr.”),
(5) Whether the approach supports regression tasks, and
(6) Whether it works with classification tasks.
We round down the values for time and memory metrics unless they are significantly greater than one. For time and memory evaluations, we assume that models are run sequentially (not in parallel) and that the entire model resides in GPU memory throughout the process. The integration metric reflects the complexity of incorporating the method into existing training pipelines or model architectures. Examples of methods with difficult integration include those requiring significant changes to model architectures (e.g., replacing most layers), training pipelines, or additional data.
Uncertainty Estimation Methods
Uncertainty estimation provides a structured way to measure how confident a deep learning model is in its predictions. These techniques play an essential role in handling noisy inputs, supporting decision-making in safety-critical settings, and identifying out-of-distribution (OOD) samples. Broadly, existing methods fall into sampling-based and sampling-free categories. Each class comes with its own advantages and limitations in terms of accuracy, computational overhead, and scalability—topics discussed in the following sections.
As discussed earlier, many uncertainty estimation techniques originate from probabilistic principles, yet most ultimately focus on producing a scalar quantity that reflects confidence in a prediction rather than modelling the full predictive distribution, which is often computationally prohibitive. In a simple 1D regression setting, for example, the model may output both a mean and an associated scale parameter such as a standard deviation [KG17]. Different approaches vary in how these quantities are obtained. The following sections outline how commonly used methods compute uncertainty and summarize them in a high-level comparison table (Table 1). Before introducing these methods, we first establish a more formal view of uncertainty from a Bayesian perspective.
Bayesian Approach
Bayesian methods offer a principled way to incorporate uncertainty into machine learning models by treating model parameters as random variables rather than fixed quantities [BBC15, GG16]. The central idea follows Bayes’ theorem, which connects the posterior distribution of the parameters \(\theta\) given observed data \(D = \{(x_i, y_i)\}_{i=1}^N\) to the prior distribution \(P(\theta)\) and the likelihood:
\[P(\theta \mid D) = \frac{P(\theta) P(D \mid \theta)}{P(D)} = \prod_{i=1}^N \frac{P(\theta) \, P(y_i \mid x_i, \theta)}{\int P(D \mid \theta) P(\theta) \, d\theta}\]where \(P(\theta \mid D)\) is the posterior distribution, representing the updated belief about the parameters after observing the data, \(P(\theta)\) is the prior, capturing prior knowledge or assumptions about \(\theta\), \(P(D \mid \theta) = \prod_{i=1}^N P(y_i \mid x_i, \theta)\) (assuming the data points are independent and identically distributed) is the likelihood, describing how probable the observed data \(D\) is given the parameters, and \(P(D)\) is the evidence, a normalizing constant ensuring that the posterior is a valid probability distribution. From this formalization, many familiar loss functions and regularization techniques in machine learning can be derived by specifying assumptions about the likelihood \(P(y_i \mid x_i, \theta)\) and the prior \(P(\theta)\). These assumptions determine the model’s behavior and the constraints imposed during training, providing a unified Bayesian perspective on various learning approaches.
In the Bayesian approach, predictions for a new input \(x\) are made by marginalizing over the posterior distribution of the parameters. The predictive distribution is expressed as:
\[P(y \mid x, D) = \int \underbrace{P(y \mid x, \theta)}_{\text{Aleatoric}} \, \underbrace{P(\theta \mid D)}_{\text{Epistemic}} \, d\theta\qquad(1)\]This equation highlights the two main sources of uncertainty in Bayesian modeling. The term \(P(y \mid x, \theta)\) represents aleatoric uncertainty, which arises from inherent noise or ambiguity in the data. It reflects the variability in predictions for a fixed model parameter \(\theta\) and is irreducible, even with additional data. In contrast, \(P(\theta \mid D)\) captures epistemic uncertainty, which stems from the model’s lack of knowledge about the true parameter values due to limited or insufficient training data. This uncertainty can be reduced by gathering more diverse or representative data, leading to a more confident posterior.
However, directly computing \(P(\theta \mid D)\) is intractable for complex models like deep neural networks due to the high-dimensional parameter space and the integral required to compute \(P(D)\). To address this, most existing uncertainty estimation methods for deep learning attempt to implicitly or explicitly approximate the Bayesian integral to make it computationally feasible. A classical strategy is to replace analytical computation with sampling. In this context, sampling from \(P(\theta \mid D)\) provides an estimate of the integral, and we refer to approaches based on this idea as sampling-based methods.
Sampling-Based Methods
Sampling-based methods derive uncertainty estimates by generating multiple predictions from the model and analyzing their variability. These approaches are known for their flexibility and accuracy, particularly in capturing epistemic uncertainty [PCH19]. As mentioned earlier, for high-dimensional real-valued parameters \(\theta\) such as those in neural networks, the integral in Equation 1 becomes intractable and is typically approximated using Monte Carlo samples:
\begin{equation} P(y \mid x, \mathcal{D}) \approx \frac{1}{N} \sum_{k=1}^{N} P(y \mid x, \theta^{(k)}); \quad \theta^{(k)} \sim p(\theta \mid \mathcal{D})\qquad(2) \label{eq:ensemble_approximation} \end{equation}
based on \(N\) different sets of parameters \(\theta^{(k)}\). Different sampling-based methods differ in how they compute these parameters. Some approaches rely on explicit posterior approximations, such as Bayesian Neural Networks [BBC15], while others use implicit sampling techniques like MC-Dropout [GG16] or Deep Ensembles [LPB17]. Although sampling-based approaches are computationally expensive due to the need to train and perform inference with multiple models, they remain the gold standard for both accuracy and uncertainty quality [OHL+19], particularly in the case of Deep Ensembles. While many sampling-free methods discussed later aim to match the performance of sampling-based approaches while reducing computational overhead, they often fall short of achieving the same level of uncertainty quality.
Masksembles: Masksembles [DDG21] enhance MC-Dropout by carefully designing structured masks that selectively drop specific features. This approach increases diversity among predictions, resulting in improved uncertainty quantification.
BatchEnsembles: BatchEnsembles [WMW20] offer an efficient ensembling approach by representing multiple models within a single network using learnable rank-one perturbations of the weights. This design minimizes memory overhead while preserving diversity in predictions. By sharing parameters across models, BatchEnsembles strike a balance between scalability and uncertainty estimation quality.
Subnetwork Inference: Subnetwork Inference [DXW21] is a scalable Bayesian framework that performs posterior inference over a small subset of neural network weights, keeping the rest as point estimates. By performing inference on a selected subnetwork with expressive approximations like full-covariance Gaussians, it aims to balance computational efficiency with accurate uncertainty estimation.
Packed-Ensembles: Packed-Ensembles [LLF22] propose an efficient ensembling approach that uses grouped convolutions to create multiple independent subnetworks within a single shared architecture. The model is trained similarly to a standard single model but produces multiple predictions, mimicking ensembling while introducing only minor computational overhead.
MIMO: The Multiple Input Multiple Output framework (MIMO) [HSG21], along with similar approaches like MixMO [rame2021mixmo], enables efficient ensembling by training independent subnetworks within a single model. During training, \(N\) inputs are processed simultaneously, allowing the model to make \(N\) independent predictions in a single forward pass. At inference, the input is replicated \(N\) times to generate \(N\) predictions, which are ensembled to estimate uncertainty.
Iterative Uncertainty: The Iterative Uncertainty framework [DDF24] leverages the iterative refinement process of neural networks to estimate uncertainty without modifying the original architecture. By analyzing the variance in outputs across consecutive iterations, it uses the rate of convergence as a proxy for certainty, where faster convergence indicates higher confidence. A related concept, Prediction Depth [BSS21], measures the layer depth at which a model’s prediction stabilizes, with harder examples requiring deeper layers for convergence.
ABNN: Adaptable Bayesian Neural Networks (ABNN) [FFM24], along with similar approaches like BatchNorm Uncertainty [ABT18], transform pre-trained DNNs into Bayesian Neural Networks by fine-tuning the model’s normalization layers and introducing noise during inference to approximate sampling.
Bayes by Backprop: Bayes by Backprop [BBC15] is a variational inference method that approximates the posterior distribution of model parameters by minimizing a variational objective. Instead of learning point estimates, it learns a distribution over weights, typically using a Gaussian variational posterior.
Sampling-Free Methods
Sampling-free methods estimate uncertainty without generating multiple predictions, offering computational efficiency and scalability. As mentioned earlier, sampling-based approaches are often too costly for real-world applications like robotics and autonomous driving, necessitating faster alternatives. These methods achieve efficiency through architectural modifications, deterministic approximations, or specialized loss functions. Several popular approaches are discussed in the following subsections.
Distance-Aware Uncertainty: Recent approaches such as Deterministic Uncertainty Quantification (DUQ) [VA20] and SNGP [LDL20] focus on single–forward-pass uncertainty estimation by introducing distance-aware output layers, such as radial basis functions (RBFs) in DUQ and Gaussian processes (GPs) in SNGP. DUQ employs a Jacobian penalty to encourage smoothness in the feature space, while SNGP uses spectral normalization to ensure sensitivity and stability in the latent space. Building on these ideas, Deep Deterministic Uncertainty (DDU) [MGE23] further extends the framework by directly modeling the feature density in the latent space.
Conformal Prediction: Conformal prediction [SW08] is a statistical framework that provides uncertainty estimates with formal coverage guarantees, ensuring that the true label lies within the predicted set with a specified confidence level. It operates by calibrating model outputs on a held-out dataset to produce prediction intervals or sets, making it agnostic to the underlying model architecture. This flexibility allows conformal prediction to be applied across a wide range of tasks, offering a principled and reliable approach to uncertainty quantification in real-world applications.
Second-Order Probabilistic Modeling: Second-order probabilistic modeling approaches—such as prior networks [MMG18], posterior networks [CPK20], and evidential learning [SEK18]—capture uncertainty by modeling distributions over predictive probabilities. These methods leverage a factorization of the predictive distribution, as shown in Equation 3, to decompose uncertainty into aleatoric, distributional, and epistemic components.
\[p(y \mid x, \mathcal{D}) = \iint \underbrace{P(y \mid \pi)}_{\text{Aleatoric}} \, \underbrace{p(\pi \mid x, \theta)}_{\text{Distributional}} \, \underbrace{p(\theta \mid \mathcal{D})}_{\text{Epistemic}} \, d\pi \, d\theta \approx \\ \approx \int \underbrace{P(y \mid \pi)}_{\text{Aleatoric}} \, \underbrace{p(\pi \mid x, \hat{\theta})}_{p(\theta \mid \mathcal{D}) \approx \delta(\theta - \hat{\theta})} \, d\pi.\qquad(3)\]Prior networks directly model the distributional uncertainty \(p(\pi \mid x, \hat{\theta})\) by parameterizing a Dirichlet distribution, enabling the network to represent confidence levels tied to predictive uncertainty. Posterior networks extend this by incorporating learned feature-level distributions, combining evidence from the data to produce refined predictions. Evidential learning follows a similar framework, interpreting input data as evidence to parameterize distributions over class probabilities, often using a Dirichlet or related family.
Additional Information Approaches: Recent methods leverage auxiliary information to probe model confidence, based on the idea that confident models remain unaffected by additional inputs, while uncertain ones show prediction shifts. ZigZag [DCF24] measures uncertainty by comparing predictions with and without auxiliary information (ground truth during training and model predictions during inference), revealing sensitivity to added data. Similarly, Experts Don’t Cheat [JGL24] leverages paired responses, allowing models to “cheat” by using one response to predict the other, thereby exposing uncertain predictions.
Conclusion
References
[LPB17] B. Lakshminarayanan, A. Pritzel, and C. Blundell. “Simple and Scalable Predictive Uncertainty Estimation Using Deep Ensembles.” In Advances in Neural Information Processing Systems, 2017.
[GG16] Y. Gal and Z. Ghahramani. “Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning.” In International Conference on Machine Learning, 2016.
[HYL+17] G. Huang, Y. Li, G. Pleiss, Z. Liu, J. E. Hopcroft, and K. Q. Weinberger. “Snapshot Ensembles: Train 1, Get M for Free.” In International Conference on Learning Representations, 2017.
[GVR+18] T. Garipov, P. Izmailov, D. Podoprikhin, T. Vetrov, and A. G. Wilson. “Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs.” In Advances in Neural Information Processing Systems, 2018.
[BBC15] C. Blundell, J. Cornebise, K. Kavukcuoglu, and D. Wierstra. “Weight Uncertainty in Neural Networks.” In International Conference on Machine Learning, 2015.
[WMW+20] Y. Wen, Y. M. Wu, E. M. Moody, and S. L. Lee. “BatchEnsemble: An Alternative Approach to Efficient Ensemble and Lifelong Learning.” In International Conference on Learning Representations, 2020.
[DDG21] N. Durasov, D. Dvornik, and P. Fua. “Masksembles for Uncertainty Estimation.” In International Conference on Computer Vision, 2021.
[LLF22] C. Laurent, S. Luneau, and J. Forest. “Packed-Ensembles for Efficient Uncertainty Estimation.” In International Conference on Learning Representations, 2022.
[HSG+21] D. Havasi, R. Senanayake, J. Liu, G. Dillon, and A. Hartnett. “Training Independent Subnetworks for Robust Prediction.” In International Conference on Learning Representations, 2021.
[DDF24] N. Durasov, D. Dvornik, and P. Fua. “Iterative Uncertainty Estimation with Self-Consistency.” In Conference on Computer Vision and Pattern Recognition, 2024.
[FFM24] G. Franchi, S. Franchi, and A. Maki. “Make Uncertainty Great Again: Accounting for Approximation in Bayesian Neural Networks.” In International Conference on Learning Representations, 2024.
[BSS+21] A. Baldock, C. Shen, A. van den Hengel, and L. Liu. “Deep Regression Ensembles with Prediction Depth.” In Advances in Neural Information Processing Systems, 2021.
[DXW21] N. Daxberger, J. Kristiadi, S. Sun, A. Immer, R. Eschenhagen, A. Pooch, and P. Hennig. “Bayesian Deep Learning via Subnetwork Inference.” In International Conference on Machine Learning, 2021.
[ABT18] A. Atanov, A. Ashukha, D. Vetrov, and M. Muhammad. “Uncertainty Estimation via Batch Normalization in Deep Networks.” In International Conference on Learning Representations, 2018.
Single-Model Predictive Uncertainty
[KG17] A. Kendall and Y. Gal. “What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?” In Advances in Neural Information Processing Systems, 2017.
[LDL20] J. Liu, C. D. Tran, A. Tagasovska, D. J. Sutherland, and I. Davidson. “Simple and Efficient Gaussian Process Priors for Deep Models.” In Advances in Neural Information Processing Systems, 2020.
[VA20] L. van Amersfoort, M. Smith, M. C. C. Labouriau, N. Pitas, and Y. Gal. “Uncertainty Estimation Using Deep Kernel Learning.” In International Conference on Machine Learning, 2020.
[MGE23] A. Mukhoti, C. Galesso, R. Guo, G. van Steenkiste, and J. M. Hernández-Lobato. “Deep Deterministic Uncertainty: A Simple Baseline.” In International Conference on Learning Representations, 2023.
[DCF24] N. Durasov, A. Charpentier, and P. Fua. “ZigZag: Adversarial Inference for OOD Detection.” In International Conference on Learning Representations, 2024.
[JGL24] A. Johnson, A. Gouk, and T. Lotte. “Experts Don’t Cheat: Reliable Deep Ensembles without Tricks.” In International Conference on Learning Representations, 2024.
[SEK18] Y. Sensoy, L. Kaplan, and M. Kandemir. “Evidential Deep Learning to Quantify Classification Uncertainty.” In Advances in Neural Information Processing Systems, 2018.
[MMG18] A. Malinin and M. Gales. “Predictive Uncertainty Estimation via Prior Networks.” In Advances in Neural Information Processing Systems, 2018.
[CPK20] A. Charpentier, S. Pionnier, and M. K. Ghassemi. “Posterior Networks: Extracting Uncertainty from Deterministic Neural Networks.” In International Conference on Learning Representations, 2020.
[PCH19] B. Postels, C. Chitta, and L. Van Gool. “Sampling-Free Epistemic Uncertainty Estimation Using Variance Propagation.” In International Conference on Computer Vision, 2019.
[TG19] A. Tagasovska and D. Gonzalez. “Orthonormal Certificates for Out-of-Distribution Detection.” In International Conference on Machine Learning, 2019.
[SW08] G. Shafer and V. Vovk. “A Tutorial on Conformal Prediction.” Journal of Machine Learning Research, 2008.