Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userRevisiting the Calibration of Modern Neural Networks
156
0
0.0
(
0
)
Added by
Matthias Minderer
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Accurate estimation of predictive uncertainty (model calibration) is essential for the safe application of neural networks. Many instances of miscalibration in modern neural networks have been reported, suggesting a trend that newer, more accurate models produce poorly calibrated predictions. Here, we revisit this question for recent state-of-the-art image classification models. We systematically relate model calibration and accuracy, and find that the most recent models, notably those not using convolutions, are among the best calibrated. Trends observed in prior model generations, such as decay of calibration with distribution shift or model size, are less pronounced in recent architectures. We also show that model size and amount of pretraining do not fully explain these differences, suggesting that architecture is a major determinant of calibration properties.
rate research
Read More
Calibrating neural networks is of utmost importance when employing them in safety-critical applications where the downstream decision making depends on the predicted probabilities. Measuring calibration error amounts to comparing two empirical distributions. In this work, we introduce a binning-free calibration measure inspired by the classical Kolmogorov-Smirnov (KS) statistical test in which the main idea is to compare the respective cumulative probability distributions. From this, by approximating the empirical cumulative distribution using a differentiable function via splines, we obtain a recalibration function, which maps the network outputs to actual (calibrated) class assignment probabilities. The spine-fitting is performed using a held-out calibration set and the obtained recalibration function is evaluated on an unseen test set. We tested our method against existing calibration approaches on various image classification datasets and our spline-based recalibration approach consistently outperforms existing methods on KS error as well as other commonly used calibration measures.
Calibration of neural networks is a critical aspect to consider when incorporating machine learning models in real-world decision-making systems where the confidence of decisions are equally important as the decisions themselves. In recent years, there is a surge of research on neural network calibration and the majority of the works can be categorized into post-hoc calibration methods, defined as methods that learn an additional function to calibrate an already trained base network. In this work, we intend to understand the post-hoc calibration methods from a theoretical point of view. Especially, it is known that minimizing Negative Log-Likelihood (NLL) will lead to a calibrated network on the training set if the global optimum is attained (Bishop, 1994). Nevertheless, it is not clear learning an additional function in a post-hoc manner would lead to calibration in the theoretical sense. To this end, we prove that even though the base network ($f$) does not lead to the global optimum of NLL, by adding additional layers ($g$) and minimizing NLL by optimizing the parameters of $g$ one can obtain a calibrated network $g circ f$. This not only provides a less stringent condition to obtain a calibrated network but also provides a theoretical justification of post-hoc calibration methods. Our experiments on various image classification benchmarks confirm the theory.
Modern neural networks have been successful in many regression-based tasks such as face recognition, facial landmark detection, and image generation. In this work, we investigate an intuitive but understudied characteristic of modern neural networks, namely, the nonsmoothness. The experiments using synthetic data confirm that such operations as ReLU and max pooling in modern neural networks lead to nonsmoothness. We quantify the nonsmoothness using a feature named the sum of the magnitude of peaks (SMP) and model the input-output relationships for building blocks of modern neural networks. Experimental results confirm that our model can accurately predict the statistical behaviors of the nonsmoothness as it propagates through such building blocks as the convolutional layer, the ReLU activation, and the max pooling layer. We envision that the nonsmoothness feature can potentially be used as a forensic tool for regression-based applications of neural networks.
Nested networks or slimmable networks are neural networks whose architectures can be adjusted instantly during testing time, e.g., based on computational constraints. Recent studies have focused on a nested dropout layer, which is able to order the nodes of a layer by importance during training, thus generating a nested set of sub-networks that are optimal for different configurations of resources. However, the dropout rate is fixed as a hyper-parameter over different layers during the whole training process. Therefore, when nodes are removed, the performance decays in a human-specified trajectory rather than in a trajectory learned from data. Another drawback is the generated sub-networks are deterministic networks without well-calibrated uncertainty. To address these two problems, we develop a Bayesian approach to nested neural networks. We propose a variational ordering unit that draws samples for nested dropout at a low cost, from a proposed Downhill distribution, which provides useful gradients to the parameters of nested dropout. Based on this approach, we design a Bayesian nested neural network that learns the order knowledge of the node distributions. In experiments, we show that the proposed approach outperforms the nested network in terms of accuracy, calibration, and out-of-domain detection in classification tasks. It also outperforms the related approach on uncertainty-critical tasks in computer vision.
Graph neural networks (GNNs) have achieved great success in recent years. Three most common applications include node classification, link prediction, and graph classification. While there is rich literature on node classification and graph classification, GNN for link prediction is relatively less studied and less understood. One common practice in previous works is to first compute node representations through a GNN, and then directly aggregate two node representations as a link representation. In this paper, we show the limitations of such an approach, and propose a labeling trick to make GNNs learn better link representations. Labeling trick assigns labels to nodes as their additional features according to nodes relationships with the target link. We show theoretically that GNNs applied to such labeled graphs can learn most expressive link representations. We also show that one state-of-the-art link prediction model, SEAL, exactly uses a labeling trick. Labeling trick brings up to 195% performance gains over plain GNNs, achieving 3 first places on the OGB link prediction leaderboard.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
