Projection Pursuit CPCANet for Domain Generalization

Hugging Face Daily Papers Papers
domain-generalization principal-component-analysis projection-pursuit orthogonal-basis stiefel-manifold deep-learning

Summary

Proposes PP-CPCANet, a covariance-free framework for domain generalization that learns a global orthogonal basis on the Stiefel manifold and achieves SOTA performance on four benchmarks.

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.
Original Article
View Cached Full Text

Cached at: 07/29/26, 11:52 AM

Paper page - Projection Pursuit CPCANet for Domain Generalization

Source: https://huggingface.co/papers/2607.22117

Abstract

DomainGeneralization(DG)aimstolearnrepresentationsrobusttodistributionshifts.Recentgeometricalignmentmethods,suchasCPCANet,extractdomain-invariantstructuresthroughbatch-wiseCommonPrincipalComponentAnalysis(CPCA).However,CPCANetsuffersfromrank-deficientcovarianceestimationduetothesmall-sample-sizeissueinmini-batchtraining.Toaddressthislimitation,weproposeProjectionPursuitCPCANet(PP-CPCANet),acovariance-freeframeworkthatlearnsaglobalorthogonalbasisontheStiefelmanifoldandjointlyoptimizesitwithnetworkparametersviatheCayleytransform.Wefurtherintroduceasymmetry-breakingdetached-medianPPdispersionobjectivetoextractcommonprincipalcomponents(CPCs)withdenseandrobustoptimizationsignals.ExperimentsonfourDGbenchmarksshowthatPP-CPCANetachievesSOTAperformancewhilemaintainingstabletraining.

View arXiv pageView PDFGitHubAdd to collection

Get this paper in your agent:

hf papers read 2607\.22117

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.22117 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.22117 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.22117 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Probe Generalization as Subspace Selection for OOD Deception Detection

arXiv cs.CL

This paper shows that projecting language model activations onto a small subset of principal components from the training distribution enables effective cross-domain transfer for deception detection probes, narrowing the performance gap between baseline and oracle methods.

A PAC-Bayesian View of Generalisation for Physics-Informed Machine Learning

arXiv cs.LG

This paper develops a PAC-Bayesian framework for physics-informed machine learning, providing high-probability generalization guarantees for unbounded losses. It proposes a multi-task perspective that jointly handles data fidelity, PDE residuals, and boundary conditions, and introduces a self-bounding learning algorithm.

Hierarchical Domain Generalization

arXiv cs.LG

This paper introduces hierarchical domain generalization, formalizing extrapolation from finite observed regions to an entire instance space. It shows that no matter how simple the hypothesis class, certain domain partitions make generalization impossible, arguing that modern generalization theory must incorporate domain structure.

Manifold-Aware General Coded Computing for Straggler-Resilient Distributed Computing

arXiv cs.LG

This paper proposes a manifold-aware encoding strategy for general coded computing that exploits the intrinsic low-dimensional geometry of high-dimensional data to improve straggler resilience in distributed systems. Experiments show reduced mean squared recovery error in neural network inference and polynomial evaluation tasks.