Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNNs) on popular benchmarks exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically salient features. When a model provides a high-confidence decision without salient supporting input features, we say the classifier has overinterpreted its input, finding too much class-evidence in patterns that appear nonsensical to humans. Here, we demonstrate that neural networks trained on CIFAR-10 and ImageNet suffer from overinterpretation, and we find models on CIFAR-10 make confident predictions even when 95% of input images are masked and humans cannot discern salient features in the remaining pixel-subsets. We introduce Batched Gradient SIS, a new method for discovering sufficient input subsets for complex datasets, and use this method to show the sufficiency of border pixels in ImageNet for training and testing. Although these patterns portend potential model fragility in real-world deployment, they are in fact valid statistical patterns of the benchmark that alone suffice to attain high test accuracy. Unlike adversarial examples, overinterpretation relies upon unmodified image pixels. We find ensembling and input dropout can each help mitigate overinterpretation.

Overinterpretation is related to overfitting, but overfitting can be diagnosed via reduced test accuracy. Overinterpretation can stem from true statistical signals in the underlying dataset distribution that happen to arise from particular properties of the data source (e.g., dermatologists' rulers).

Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNNs) on popular benchmarks exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically salient features. When a model provides a high-confidence decision without salient supporting input features, we say the classifier has overinterpreted its input, finding too much class-evidence in patterns that appear nonsensical to humans. Here, we demonstrate that neural networks trained on CIFAR-10 and ImageNet suffer from overinterpretation, and we find models on CIFAR-10 make confident predictions even when 95% of input images are masked and humans cannot discern salient features in the remaining pixel-subsets. We introduce Batched Gradient SIS, a new method for discovering sufficient input subsets for complex datasets, and use this method to show the sufficiency of border pixels in ImageNet for training and testing.

Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNNs) on popular benchmarks exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically salient features. When a model provides a high-confidence decision without salient supporting input features, we say the classifier has overinterpreted its input, finding too much class-evidence in patterns that appear nonsensical to humans. Here, we demonstrate that neural networks trained on CIFAR-10 and ImageNet suffer from overinterpretation, and we find models on CIFAR-10 make confident predictions even when 95% of input images are masked and humans cannot discern salient features in the remaining pixel-subsets. We introduce Batched Gradient SIS, a new method for discovering sufficient input subsets for complex datasets, and use this method to show the sufficiency of border pixels in ImageNet for training and testing.

Scientists at the Massachusetts Institute of Technology have stumbled upon an interesting problem with machine learning and image classification. This problem, if not solved, could be harmless or could be deadly, depending on what the system is being used for. Simply put, a model could look at an image and make a prediction based on information that we humans can't make sense of, and it could be wrong. Image classification is used in both medical diagnostics and autonomous driving. The aim is to train a neural network to understand an image in a similar way that a human does.

For all that neural networks can accomplish, we still don't really understand how they operate. Sure, we can program them to learn, but making sense of a machine's decision-making process remains much like a fancy puzzle with a dizzying, complex pattern where plenty of integral pieces have yet to be fitted. If a model was trying to classify an image of said puzzle, for example, it could encounter well-known, but annoying adversarial attacks, or even more run-of-the-mill data or processing issues. But a new, more subtle type of failure recently identified by MIT scientists is another cause for concern: "overinterpretation," where algorithms make confident predictions based on details that don't make sense to humans, like random patterns or image borders. This could be particularly worrisome for high-stakes environments, like split-second decisions for self-driving cars, and medical diagnostics for diseases that need more immediate attention.

For all that neural networks can accomplish, we still don't really understand how they operate. Sure, we can program them to learn, but making sense of a machine's decision-making process remains much like a fancy puzzle with a dizzying, complex pattern where plenty of integral pieces have yet to be fitted. If a model was trying to classify an image of said puzzle, for example, it could encounter well-known, but annoying adversarial attacks, or even more run-of-the-mill data or processing issues. But a new, more subtle type of failure recently identified by MIT scientists is another cause for concern: "overinterpretation," where algorithms make confident predictions based on details that don't make sense to humans, like random patterns or image borders. This could be particularly worrisome for high-stakes environments, like split-second decisions for self-driving cars, and medical diagnostics for diseases that need more immediate attention.

Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNN) exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically salient features. When a model provides a high-confidence decision without salient supporting input features we say that the classifier has overinterpreted its input, finding too much class-evidence in patterns that appear nonsensical to humans. Here, we demonstrate that state of the art neural networks for CIFAR-10 and ImageNet suffer from overinterpretation, and find CIFAR-10 trained models make confident predictions even when 95% of an input image has been masked and humans are unable to discern salient features in the remaining pixel subset. Although these patterns portend potential model fragility in real-world deployment, they are in fact valid statistical patterns of the image classification benchmark that alone suffice to attain high test accuracy. We find that ensembling strategies can help mitigate model overinterpretation, and classifiers which rely on more semantically meaningful features can improve accuracy over both the test set and out-of-distribution images from a different source than the training data.