Light-FMP: Lightweight Feature and Model Pruning for Enhanced Deep Recommender Systems
Nghia Bui, Yue Ning, Lijing Wang
TLDR
Light-FMP is a lightweight framework for deep recommender systems that prunes features and models to enhance both computational efficiency and accuracy.
Key contributions
- Uses a hard concrete distribution for efficient pretraining to identify important features.
- Prunes both model parameters and identified features to reduce computational overhead.
- Continues training on remaining data with domain-adapted parameters for optimal performance.
- Achieves superior efficiency and accuracy compared to existing deep recommender system methods.
Why it matters
Deep recommender systems struggle to balance efficiency and accuracy, especially with high-dimensional data. Light-FMP offers a novel solution by simultaneously optimizing both, leading to more practical and performant systems. This advances DRS by providing a scalable and robust approach.
Original Abstract
Deep recommender systems (DRS) often face challenges in balancing computational efficiency and model accuracy, especially when handling high-dimensional input features. Existing methods either focus on improving accuracy while neglecting training efficiency or prioritize efficiency at the cost of suboptimal accuracy across tasks. We propose Light-FMP: Lightweight Feature and Model Pruning for Enhanced DRS, a lightweight framework that addresses the challenges through three key phases: \textit{pretraining}, \textit{pruning}, and \textit{continued training}. Using a hard concrete distribution, a masking layer is efficiently pretrained on a small data subset to identify important features. The model and features are then pruned, and training continues on the remaining dataset with domain-adapted parameters. Experiments on benchmark datasets from real-world recommender systems demonstrate that Light-FMP outperforms existing methods in both efficiency and accuracy while maintaining scalability and robustness.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.