Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Projection Pursuit CPCANet (PP-CPCANet), a novel framework designed to enhance Domain Generalization (DG) by addressing the limitations of existing methods like CPCANet, which struggle with rank-deficient covariance estimation due to small sample sizes. By leveraging a covariance-free approach that learns a global orthogonal basis on the Stiefel manifold and optimizing it alongside network parameters using the Cayley transform, PP-CPCANet effectively extracts common principal components with improved robustness. Experimental results across four DG benchmarks demonstrate that PP-CPCANet achieves state-of-the-art performance while ensuring stable training dynamics.
Achieving state-of-the-art performance in domain generalization, PP-CPCANet overcomes the small-sample-size limitations of existing methods with a novel covariance-free approach.
Domain Generalization (DG) aims to learn representations robust to distribution shifts. Recent geometric alignment methods, such as CPCANet, extract domain-invariant structures through batch-wise Common Principal Component Analysis (CPCA). However, CPCANet suffers from rank-deficient covariance estimation due to the small-sample-size issue in mini-batch training. To address this limitation, we propose Projection Pursuit CPCANet (PP-CPCANet), a covariance-free framework that learns a global orthogonal basis on the Stiefel manifold and jointly optimizes it with network parameters via the Cayley transform. We further introduce a symmetry-breaking detached-median PP dispersion objective to extract common principal components (CPCs) with dense and robust optimization signals. Experiments on four DG benchmarks show that PP-CPCANet achieves SOTA performance while maintaining stable training.