Slumbering to Precision: Enhancing Artificial Neural Network Calibration Through Sleep-like Processes
Authors: Jean Erik Delanois, Aditya Ahuja, Giri P. Krishnan, Maxim Bazhenov
Organizations: Department of Computer Science & Engineering, University of California, San Diego, La Jolla, California, USA · Department of Medicine, University of California, San Diego, La Jolla, California, USA · ARTISAN, Georgia Institute of Technology, Atlanta, Georgia, USA
Artificial neural networks are often overconfident, undermining trust because their predicted probabilities do not match actual accuracy. Inspired by biological sleep and the role of spontaneous replay in memory and learning, we introduce Sleep Replay Consolidation (SRC), a novel calibration approach. SRC is a post-training, sleep-like phase that selectively replays internal representations to update network weights and improve calibration without supervised retraining. Across multiple experiments, SRC is competitive with and complementary to standard approaches such as temperature scaling. Combining SRC with temperature scaling achieves the best Brier score and entropy trade-offs for AlexNet and VGG19. These results show that SRC provides a fundamentally novel approach to improving neural network calibration. SRC-based calibration offers a practical path toward more trustworthy confidence estimates and narrows the gap between human-like uncertainty handling and modern deep networks.
Figures & tables
Accuracy
ECE
NLL
Brier
Entropy
ResNet 152 on CIFAR 100
Baseline
84.9
0.062184
0.601612
9.369636
0.457585
Baseline + SRC
84.8
0.012978
0.570327
4.33612
1.015678
Baseline + TS
84.9
0.016904
0.562992
5.478358
0.838229
Baseline + SRC + TS
84.8
0.014159
0.565073
4.940667
0.857810
Baseline + LS
85.2 †
0.026636
0.617318
1.624467 †
1.162224 †
Table 1: Accuracy and calibration metrics averaged over 10 trials (STD in Appendix B). Bold values indicate the best-performing post-hoc method. Bold values marked with ( † ) indicate methods that achieved the best score but required retraining.
Figure 1: Reliability diagrams showing improvement in ECE metric as a result of SRC (Right) after initial Baseline training (Left) for the ResNet-152 model trained on CIFAR-100. Confidence more accurately reflects accuracy (blue bins closer to diagonal). Red bars indicate difference between ideal and actual accuracy. Test confidence and accuracy are shown.
Figure 2: Two-dimensional histograms of ResNet-152 confidences on CIFAR-100 baseline confidence (horizontal) vs method confidence (vertical); color encodes sample density and the red diagonal marks no change. SRC (left) increases or decreases confidences; TS (right) only maintains or reduces them.
Figure 3: Histograms of FF feature magnitudes for ResNet-152 on CIFAR-100: Baseline/TS (blue), SRC (orange), and LS (green). The Baseline model shows the widest distribution (0–8), LS is more constrained (0–4.5), and SRC shifts the Baseline distribution to a lower maximum (5.5), closely matching LS. This suggests SRC aligns feature representations with the well-calibrated LS model without full retraining.
Figure 4: Distributions of nonzero FF layer elements over test samples in ResNet-152 on CIFAR-100 for Baseline/TS (blue), SRC (orange), and LS (green). Baseline representations are densest (82–90% nonzero) and LS is moderately sparser with partial overlap. SRC achieves substantially sparser representations (67–81%).
Figure 5: Weight changes from SRC for ResNet-152 on CIFAR-100. Most weights decreased, leading to smaller feature magnitudes (similar to LS) and sparser representations.
Neural networks (NNs), while highly flexible and versatile, are often poorly calibrated. Post-hoc calibration methods adjust the outputs of an already trained model using a separate calibration dataset and additional calibration parameters. This process adds overhead that requires separate optimization of the calibrator and subsequent model selection. In this paper, we show that standard (post-hoc) calibrators can be trained end-to-end with the backbone via a bilevel Stackelberg formulation. This formulation eliminates the need for a separate calibration set or a dedicated calibration loss by jointly optimizing accuracy and calibration using cross-entropy, introducing only one additional hyperparameter: the calibrator learning rate. Although our framework supports calibrators ranging from temperature and matrix scaling to NN layers, our results show that using the final classification layer itself as the calibrator yields accurate, well-calibrated models with no additional parameters. Our approach outperforms SOTA calibrators with minimal added complexity, while also avoiding the multiplicative overhead of separate calibration components in settings such as multi-task learning.
Nektarios Kalampalikis, Pablo M. Olmos, Isabel Valera
Dept. of Computer Science Saarland University, Germany
Although deep neural networks (DNNs) achieve high predictive accuracy, their confidence estimates are often unreliable, potentially compromising user trust in their decisions. This has motivated research on calibrated models, where calibration measures how well a model's predicted confidence aligns with the empirical probability of correctness. However, calibration metrics can often be improved through post-processing techniques that merely mimic training-time uncertainty without genuinely improving the model's understanding. For this reason, statisticians recommend that models be not only calibrated but also refined. Intuitively, a model is considered more refined if it assigns significantly different confidence scores to correct and incorrect predictions, a property also referred to as sharpness. We observe that many existing calibration methods improve calibration at the cost of reduced refinement. To address this limitation, we propose: (1) a novel loss function that explicitly promotes refinement and can be optimized through supervised contrastive learning; and (2) a unified training framework, RefCal, that jointly optimizes calibration, refinement, and accuracy to improve DNN reliability. On the CIFAR-100-LT dataset with 10 percent class imbalance, RefCal achieves (accuracy, refinement, ECE) of (58.81, 95.67, 0.08), substantially outperforming the widely used Correctness Ranking Loss, which achieves (46.27, 93.7, 0.22).
Modern neural networks are often miscalibrated, with a tendency to overconfidence. Existing train-time calibration methods largely modify task losses or calibration penalties, leaving neighborhood structure in learned representations underexploited. We introduce graph smoothing as a general principle for train-time calibration, which encourages similar predictive distributions across neighboring samples in representation space. We analyze the effects of graph smoothing, deriving bounds that connect predictive divergence between neighboring samples to local confidence variation and to the propagation of pointwise calibration error, and characterize the conditions under which smoothing can or cannot improve calibration. In light of this analysis, we propose \modelNoSpace, a graph-based train-time regularizer that penalizes the Jensen--Shannon divergence between predictive distributions of neighboring samples. We present a thorough empirical analysis, showing that across standard calibration benchmarks, \model improves predictive quality, and the improvement is complementary to post-hoc calibration: after temperature scaling, \model attains the lowest NLL of all evaluated train-time methods in seven of the eight image and tabular settings. These findings demonstrate the value of graph smoothing over learned representations for neural network calibration.
Idan Horowitz, Avigdor Gal
Faculty of Data and Decision Sciences Technion Haifa, 3200003, IL