Approach to Finding a Robust Deep Learning Model

2025-05-22

AI Summary

Unsupervised large-scale model training suffers from low prediction reliability and difficulty in robustness assessment. Method: This paper proposes a fully automated, task-agnostic framework for quantifying deep learning model robustness, accompanied by a meta-level model selection algorithm. Leveraging a lightweight CNN-FC hybrid architecture, we systematically investigate the coupled effects of sample size, weight initialization, and inductive bias on robustness. The meta-selection algorithm is agnostic to both model architecture and downstream tasks. Results: Empirical analysis reveals that initialization strategy and data scale are primary determinants of robustness. Our approach substantially improves prediction reliability and enables efficient, reproducible robustness ranking—even for lightweight models—thereby establishing a new paradigm for automated model evaluation and deployment.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

The rapid development of machine learning (ML) and artificial intelligence (AI) applications requires the training of large numbers of models. This growing demand highlights the importance of training models without human supervision, while ensuring that their predictions are reliable. In response to this need, we propose a novel approach for determining model robustness. This approach, supplemented with a proposed model selection algorithm designed as a meta-algorithm, is versatile and applicable to any machine learning model, provided that it is appropriate for the task at hand. This study demonstrates the application of our approach to evaluate the robustness of deep learning models. To this end, we study small models composed of a few convolutional and fully connected layers, using common optimizers due to their ease of interpretation and computational efficiency. Within this framework, we address the influence of training sample size, model weight initialization, and inductive bias on the robustness of deep learning models.

Problem

• Ensuring reliable predictions in unsupervised deep learning models
• Developing a versatile robustness evaluation approach for ML models
• Assessing impact of training size and initialization on model robustness

Innovation

• Meta-algorithm for model selection
• Unsupervised robust model training
• Evaluates robustness via sample size