Do CIFAR-10 Classifiers Generalize to CIFAR-10?
read the original abstract
Machine learning is currently dominated by largely experimental work focused on improvements in a few key tasks. However, the impressive accuracy numbers of the best performing models are questionable because the same test sets have been used to select these models for multiple years now. To understand the danger of overfitting, we measure the accuracy of CIFAR-10 classifiers by creating a new test set of truly unseen images. Although we ensure that the new test set is as close to the original data distribution as possible, we find a large drop in accuracy (4% to 10%) for a broad range of deep learning models. Yet more recent models with higher original accuracy show a smaller drop and better overall performance, indicating that this drop is likely not due to overfitting based on adaptivity. Instead, we view our results as evidence that current accuracy numbers are brittle and susceptible to even minute natural variations in the data distribution.
This paper has not been read by Pith yet.
Forward citations
Cited by 10 Pith papers
-
How does the optimizer implicitly bias the model merging loss landscape?
Effective noise scale non-monotonically governs model merging success with an optimum, unifying effects of learning rate, weight decay, batch size, and augmentation on the loss landscape.
-
Convex Optimization for Alignment and Preference Learning on a Single GPU
COALA applies convex optimization reformulations of neural networks to direct preference optimization, claiming single-GPU training with ~18% of DPO's TFLOPs and competitive performance on multiple datasets and models...
-
Encrypted Neural Networks without Overflows
Introduces formal verification to compute certified neuron range bounds for CKKS-encrypted neural networks, eliminating overflow failures that previously reached 47%.
-
TopoGeoScore: A Self-Supervised Source-Only Geometric Framework for OOD Checkpoint Selection
TopoGeoScore combines a torsion-inspired Laplacian log-determinant, Ollivier-Ricci curvature, and higher-order topological summaries from source embeddings, with weights learned via self-supervised invariance to geome...
-
A Comparative Analysis on the Performance of Upper Confidence Bound Algorithms in Adaptive Deep Neural Networks
UCB-V and UCB-Tuned dominate accuracy-energy and accuracy-latency trade-offs while all tested UCB strategies achieve sub-linear regret in adaptive DNN early-exit experiments on CIFAR datasets.
-
Tent: Fully Test-time Adaptation by Entropy Minimization
Test-time entropy minimization adapts models by optimizing for confident predictions, reducing error on corrupted ImageNet-C and enabling source-free domain adaptation.
-
A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark
VTAB is a 19-task benchmark that measures representation quality by few-shot adaptation performance across diverse vision domains, with a controlled large-scale comparison of popular pretraining methods.
-
A Comparative Analysis on the Performance of Upper Confidence Bound Algorithms in Adaptive Deep Neural Networks
Comparative study applies UCB-V, UCB-Tuned, UCB-Bayes and UCB-BwK to ADNN early-exit selection on ResNet and MobileViT using CIFAR-10/100, reporting sub-linear regret with UCB-Bayes fastest and UCB-V/UCB-Tuned best on...
-
Generative Channel Knowledge Base With Environmental Information for Joint Source-Channel Coding in Semantic Communications
A Transformer-based generative model builds an environment-aware channel knowledge base that is injected into JSCC encoders and decoders, achieving 10^{-3} level channel estimation error and outperforming benchmarks i...
-
Reproducibility in Machine Learning for Health
Systematic evaluation of over 100 ML4H papers finds poorer reproducibility than other ML fields, driven by limited data and code access, and offers recommendations to data providers, publishers, and researchers.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.