Software
Real-time streaming and processing of brain signals inside MNE-Python, over any LSL-compatible device. Published in the Journal of Open Source Software, 2025, and maintained with the MNE community.
A real-time neural decoding framework covering acquisition, training, online classification, visualization and feedback. Written for the exoskeleton experiments, and later listed among the FCBG platforms at Campus Biotech. It runs the decoding in the exoskeleton work [here], the subthalamic decoding work [here], and the adaptive stimulation trial [here].
Microsoft decoding challenge solution
Gradient boosting on spectral and time-domain features of intracranial recordings, written for the 2016 Microsoft Brain Signal Decoding Challenge, where it took First Prize with the second-highest accuracy (92.5%) on the private test set. Runs on plain Python or in Azure ML Studio.
Stochastic context-free grammar learner
Learns grammar structure and parameters from noisy symbol sequences, so a robot can recover the hierarchy of a human task from a handful of demonstrations.
Data
Temporally structured human activity dataset
Six structured activities, each a sequence of action components, with several actors working in parallel in full HD. Built for benchmarking concurrent activity detection, and used in the attention-relocation work [here]. Funded by a Pump-priming Vision and Language grant that I received from the EPSRC (GBP 2,000, 2013) during my PhD.