Introducing GMD-AI: Why we started this group

Why GMD-AI?

AI systems are advancing faster than researchers can fully understand, verify, and secure them. Language models can produce convincing human-like text, while image, audio, and video generators make synthetic media harder to distinguish from authentic recordings. At the same time, adversarial attacks expose weaknesses in systems already in use.

We founded the General Machine Intelligence for Discovery, Detection, Defense (GMD-AI) research group because these problems affect one another. New capabilities create new opportunities for misuse. Detection research reveals where systems fail, and those findings guide stronger defenses. Working across all three areas helps us study the full chain from capability to risk and response.

Our three research areas

Discovery

We study modular architectures that adapt to new tasks without forgetting earlier ones, along with federated systems that train across institutions without exposing raw data. This work includes continual learning, mixture-of-experts models, and distributed optimization, with a focus on learning incrementally across multiple sites.

Detection

As generative AI improves, synthetic media becomes harder to identify. We work on deepfake detection, AI-generated content identification, and generation-time watermarking. Our tests examine whether detectors work across generators, survive post-processing, and hold up under adversarial manipulation.

Defense

New capabilities also create new ways to attack AI systems. Adversarial examples can fool classifiers, data poisoning can corrupt training pipelines, and model extraction can expose intellectual property. We study adversarial machine learning, robustness certification, and secure federated training protocols. The goal is to address individual vulnerabilities and design systems that can withstand attacks from the start.

Who we are

GMD-AI is a collaborative research group currently based across two institutions in Ireland:

• Van-Tuan Tran (Trinity College Dublin), researching modular deep learning, federated learning, and continual learning.
• Hong-Hanh Nguyen-Le (University College Dublin), researching deepfake detection, AI security, adversarial ML, and federated learning.

We publish our research openly so others can examine, reproduce, and build on it.

What to expect

This site brings together our peer-reviewed papers and preprints, current projects, open-source software, and research notes. We publish code on GitHub, papers on arXiv, and use the blog for tutorials, technical notes, and updates.

Get involved

Researchers and practitioners working on related problems can reach us through the contact page or our institutional profiles.