MNIST-Gen: A Modular MNIST-Style Dataset Generation Using Hierarchical Semantics, Reinforcement Learning, and Category Theory

2025-07-15

AI Summary

Existing MNIST-style datasets lack domain specificity (e.g., tree species, food items), while constructing custom datasets manually incurs high costs and regulatory compliance risks. Method: We propose an automated framework for generating user-customized MNIST-style image datasets. It employs a hierarchical semantic generation architecture formalized via category theory to ensure modularity and composability; integrates CLIP-based semantic guidance, reinforcement learning optimization, and human-in-the-loop feedback; and supports three generation modes—individual review, intelligent batch, and rapid batch. Contribution/Results: This work pioneers the integration of formal semantic modeling with multimodal generative methods, substantially reducing manual intervention. We successfully construct Tree-MNIST and Food-MNIST—two novel domain-specific benchmarks—with automatic classification accuracy reaching 85% and generation time reduced by 80% versus manual annotation. The framework demonstrates significant advances in domain adaptability, scalability, and practical utility.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Multimodal Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

Neural networks are often benchmarked using standard datasets such as MNIST, FashionMNIST, or other variants of MNIST, which, while accessible, are limited to generic classes such as digits or clothing items. For researchers working on domain-specific tasks, such as classifying trees, food items, or other real-world objects, these data sets are insufficient and irrelevant. Additionally, creating and publishing a custom dataset can be time consuming, legally constrained, or beyond the scope of individual projects. We present MNIST-Gen, an automated, modular, and adaptive framework for generating MNIST-style image datasets tailored to user-specified categories using hierarchical semantic categorization. The system combines CLIP-based semantic understanding with reinforcement learning and human feedback to achieve intelligent categorization with minimal manual intervention. Our hierarchical approach supports complex category structures with semantic characteristics, enabling fine-grained subcategorization and multiple processing modes: individual review for maximum control, smart batch processing for large datasets, and fast batch processing for rapid creation. Inspired by category theory, MNIST-Gen models each data transformation stage as a composable morphism, enhancing clarity, modularity, and extensibility. As proof of concept, we generate and benchmark two novel datasets- extit{Tree-MNIST} and extit{Food-MNIST}-demonstrating MNIST-Gen's utility for producing task-specific evaluation data while achieving 85% automatic categorization accuracy and 80% time savings compared to manual approaches.

Problem

• Generates custom MNIST-style datasets for domain-specific tasks
• Automates dataset creation with minimal manual intervention
• Enhances modularity and extensibility using category theory

Innovation

• Modular MNIST-style dataset generation framework
• Combines CLIP, reinforcement learning, human feedback
• Category theory models composable data transformations