Anthony Hughes

Advances in artificial intelligence have led to powerful large language models (LLMs) and large multimodal models (MMs), enabling new ways to process sensitive and complex information. However, their reliance on static, centralized data raises concerns regarding privacy, adaptability, and long-term alignment with dynamic environments.

My research focuses on building AI systems that can continually learn and adapt over time, minimizing catastrophic forgetting while maintaining privacy and security. I explore decentralized and federated learning approaches, allowing models to improve collaboratively across distributed networks while respecting data privacy constraints. These methods are particularly crucial in sensitive domains like healthcare, where ensuring security and trust in AI-driven decision-making is paramount.

In healthcare, multimodal AI systems must handle a diverse range of data. My research investigates how autonomous AI agents can operate within sensitive settings, learning from multiple sources while remaining robust to shifts in data and tasks. By fostering collaborative intelligence, we aim to create AI systems that are interpretable, adaptable, and aligned with real-world needs without compromising individual privacy.