48: Feature Selection for Latent Factor Models

Music City Center 

The research presents a comprehensive overview of feature selection methodologies in machine learning, addressing the challenges posed by high-dimensional data and the need to mitigate the curse of dimensionality. The project is built upon improving model performance via features section methods. Various approaches for feature selection, including supervised and unsupervised methods, have been outlined, and new strategies have been proposed to introduce robustness and sparsity in the feature selection process. Furthermore, it highlights the importance of evaluating these methods within a multi-class classification framework using simulated and real-world datasets. The study's contributions include introducing an SNR-based feature selection technique, exploring feature recovery guarantees, proposing robust methods for outlier handling, incorporating per-class feature selection for multi-class classification, and conducting extensive experiments to validate the proposed methods' efficacy and robustness.

Feature Selection

Latent Factor Models

Robust Loss Optimization

Multiclass Classification

Signal-to-Noise Ratio(SNR) 

Section on Statistical Learning and Data Science