Aaqib Saeed

I build machine learning systems that learn from messy real-world data such as audio, biosignals, sensor streams and text, with little supervision, and that run where the data already lives.

Assistant Professor at Eindhoven University of Technology, where I lead the Self-Learning Systems (SLS) Lab. Previously Google Research, the University of Cambridge, and Philips Research. My research is funded by an NGF AiNed Fellowship.

Research

Our work centers on self-learning foundation models that perceive signals the way experts do, act on what they find, and keep improving from their own feedback instead of waiting for more human labels.

Multimodal & foundation models

Foundation models that place audio, biosignals, sensor streams and text in one shared representation. Post-training then turns them into zero- and few-shot learners for clinical and everyday sensing, where labels are scarce.

AcuLa · D-BETA · ZETA · CardioState-JEPA

Agents & self-improving systems

Agents that plan, write code, call tools and critique one another. We treat prompts, programs and strategies as text to be optimized with feedback, so a system gets better without retraining the model underneath.

Helmsman · PRISM · StethoLM · TRIAGE

AI for science

Foundation models that propose hypotheses, build simulators and mechanistic models, and refine them against scarce real data, turning what a model knows into scientific understanding that can be inspected, tested and reused.

SimAuthor

Decentralized & efficient learning

Training across devices under noisy labels, heterogeneous sensors and tight communication budgets, and inference small enough to run on the hardware that is actually there.

FAST · FedNS · DeltaMask · FedSTAR

Publications

Selected work

Full list on Google Scholar.

Group

Self-Learning Systems (SLS) Lab

PhD researchers in the lab at Eindhoven University of Technology.

Haoyuan LiFederated learning, LLM agents for ML systems Tsai-Ning WangAudio understanding, post-training alignment
Ran PiaoCross-modal representation learning, audio language models
Mustafa IlerisoySafety of multimodal audio language models Hamza ShafiqueFoundation multimodal biosignal models
Yishan WangAgentic AI, multimodal learning

Alumni

Jialu TangPhD 2023–2026, now Postdoc, IE&IS, TU/e

Co-supervision

PhD students supervised jointly with colleagues outside the lab.

  • Daan Janssen — co-supervised with prof. dr. Hareld Kemps.
  • Francesco Moschella — co-supervised with dr. Steven Houben.

Joining

Openings are announced through TU/e and on LinkedIn. MSc students looking for a thesis project, and researchers interested in a visit or a joint proposal, are welcome to write. Please include something you have built.

Contact

Email is the best way to reach me. I collaborate with research groups and industry teams, and take on a small number of advisory engagements.