Researchers introduced NeuroStrata, a connectivity-aware deep representation learning framework for EEG-based mental stress analysis, achieving a peak accuracy of 97.3% using the LAION-CLIP-ViT-L14 ba…
A new study from arXiv (arXiv:2608.00135v1) evaluating Convolutional Neural Networks on the JONES-19 cultural design dataset finds that learning from scratch with repeated local sampling (multi-crop) …
Researchers propose a variation of Fidelity-based XAI metrics for evaluating explanations of CNN classifiers in real-world applications with few classes. The method generates in-distribution, uncertai…
Researchers from an undisclosed institution introduced Convolutional Nearest Neighbors (ConvNN), a unified framework that shows convolution and self-attention are special cases of k-nearest neighbor a…
A developer conducted controlled experiments on CIFAR-10 using a Convolutional Neural Network and found that data preprocessing alone caused model accuracy to range from 65% to over 87%, with one case…