ContextGPT combines large language models with neuro-symbolic AI to improve human activity recognition systems with less labeled data.

• Integrates common-sense knowledge from LLMs into activity recognition models
• Addresses the critical challenge of labeled data scarcity
• Demonstrates superior performance compared to traditional supervised learning approaches
• Enables more effective context-aware security monitoring

For security applications, this approach allows more reliable detection of suspicious behaviors, unauthorized activities, and potential threats while requiring significantly less training data—making advanced surveillance systems more practical to deploy and maintain.

Original Paper: ContextGPT: Infusing LLMs Knowledge into Neuro-Symbolic Activity Recognition Models