Deep neural networks exhibit distributed high-level semantic representations and discontinuous input-output mappings vulnerable to transferable adversarial perturbations.
hub
Visualizing and Understanding Convolutional Networks
15 Pith papers cite this work, alongside 450 external citations. Polarity classification is still indexing.
15
Pith papers citing it
450
external citations · Pith
abstract
Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we address both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. We also perform an ablation study to discover the performance contribution from different model layers. This enables us to find model architectures that outperform Krizhevsky \etal on the ImageNet classification benchmark. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.
hub tools
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
Gradient ascent on class scores and input-image gradients produce visualizations of ConvNet class notions and saliency maps usable for weakly supervised segmentation.
Prophecy infers formal properties of feed-forward neural networks by extracting rules from neuron activation patterns that imply desirable output behaviors.
A Shapley-value method with interaction terms that explains object detector decisions by capturing collective pixel contributions for localization and classification.
Circuit-based metrics from Vision Transformer internals provide better label-free proxies for generalization under distribution shift than existing methods like model confidence.
Introduces a perturbation-based fidelity metric tailored to few-class CNN classifiers for real-conditions XAI evaluation, tested on medical and natural imaging against human-centric metrics.
NRM enables OoD detection by joint latent likelihood, assigning lower values to SVHN than CIFAR-10 (unlike VAEs/flows) and consistent across other OoD sets.
Shapley value and variational importance switch methods produce consistent rankings of filter importance in CNNs, enabling compression and interpretability.
Modified feedback alignment in convolutional networks produces representations geometrically aligned with backpropagation on CIFAR-10.
MEFA enables exact full-gradient white-box attacks on iterative stochastic purification defenses like diffusion and Langevin EBMs by trading recomputation for lower memory, revealing vulnerabilities missed by approximate-gradient methods.
IViT applies quadratic programming to a pre-trained Vision Transformer with a multi-objective loss, achieving 93.80% accuracy on six skin disease datasets (0.21% below baseline) while reducing feature redundancy by 29.5% and producing clinically consistent activations.
Convolutional neural network decoders achieve good performance on surface code error correction and adapt across noise models, with explainable AI used to inspect their decisions.
Deep recurrent autoencoders convert images to shortened audio signals that incorporate hearing models, enabling above-chance hand posture discrimination and object reaching after a few hours of training instead of months.
Experiments indicate deep learning models achieve higher accuracy on numerosity tasks for counts below human subitizing capacity.
RNN and LSTM models outperform other algorithms in predicting stream flow from precipitation, land use, and temperature, with a public dataset released.
citing papers explorer
-
Intriguing properties of neural networks
Deep neural networks exhibit distributed high-level semantic representations and discontinuous input-output mappings vulnerable to transferable adversarial perturbations.
-
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
Gradient ascent on class scores and input-image gradients produce visualizations of ConvNet class notions and saliency maps usable for weakly supervised segmentation.
-
Prophecy: Inferring Formal Properties from Neuron Activations
Prophecy infers formal properties of feed-forward neural networks by extracting rules from neuron activation patterns that imply desirable output behaviors.
-
Explaining Object Detectors via Collective Contribution of Pixels
A Shapley-value method with interaction terms that explains object detector decisions by capturing collective pixel contributions for localization and classification.
-
Inside-Out: Measuring Generalization in Vision Transformers Through Inner Workings
Circuit-based metrics from Vision Transformer internals provide better label-free proxies for generalization under distribution shift than existing methods like model confidence.
-
Few-class Fidelity: Evaluating Explanations of Real-conditions CNN classifiers with Optimized Perturbations
Introduces a perturbation-based fidelity metric tailored to few-class CNN classifiers for real-conditions XAI evaluation, tested on medical and natural imaging against human-centric metrics.
-
Out-of-Distribution Detection Using Neural Rendering Generative Models
NRM enables OoD detection by joint latent likelihood, assigning lower values to SVHN than CIFAR-10 (unlike VAEs/flows) and consistent across other OoD sets.
-
Neuron ranking -- an informed way to condense convolutional neural networks architecture
Shapley value and variational importance switch methods produce consistent rankings of filter importance in CNNs, enabling compression and interpretability.
-
Biological Plausibility and Representational Alignment of Feedback Alignment in Convolutional Networks
Modified feedback alignment in convolutional networks produces representations geometrically aligned with backpropagation on CIFAR-10.
-
Memory Efficient Full-gradient Attacks (MEFA) Framework for Adversarial Defense Evaluations
MEFA enables exact full-gradient white-box attacks on iterative stochastic purification defenses like diffusion and Langevin EBMs by trading recomputation for lower memory, revealing vulnerabilities missed by approximate-gradient methods.
-
IViT: A Novel Interpretable Visual Transformer for Skin Disease Detection
IViT applies quadratic programming to a pre-trained Vision Transformer with a multi-objective loss, achieving 93.80% accuracy on six skin disease datasets (0.21% below baseline) while reducing feature redundancy by 29.5% and producing clinically consistent activations.
-
Convolutional neural network based decoders for surface codes
Convolutional neural network decoders achieve good performance on surface code error correction and adapt across noise models, with explainable AI used to inspect their decisions.
-
Autoencoding sensory substitution
Deep recurrent autoencoders convert images to shortened audio signals that incorporate hearing models, enabling above-chance hand posture discrimination and object reaching after a few hours of training instead of months.
-
On the notion of number in humans and machines
Experiments indicate deep learning models achieve higher accuracy on numerosity tasks for counts below human subitizing capacity.
-
Water Preservation in Soan River Basin using Deep Learning Techniques
RNN and LSTM models outperform other algorithms in predicting stream flow from precipitation, land use, and temperature, with a public dataset released.