<ul><li>Out-of-Distribution (OoD) detection is crucial for ensuring the reliability of deep neural networks by distinguishing between OoD and In-Distribution (InD) data.</li><li>This study focuses on leveraging a discriminative non-linear subspace learned from InD features to capture patterns specific to InD data, enabling effective OoD detection.</li><li>The Kernel Principal Component Analysis (KPCA) framework is utilized to create the non-linear subspace, and the reconstruction error on this subspace is used to differentiate between InD and OoD data.</li><li>The challenges addressed in the study include selecting an optimal kernel function for KPCA and developing efficient techniques to compute the kernel matrix with large-scale InD data, leading to a KPCA detection method with enhanced efficacy and efficiency.</li></ul>

Kernel PCA for Out-of-Distribution Detection: Non-Linear Kernel Selections and Approximations

Discover more