Teaching

Joe Francis teaches two cross-listed course pairs in the Department of Biomedical Engineering, each open to senior undergraduates and graduate students. Both spend real class time on ethics, rigor and reproducibility.

Brain Machine Interfacing

BIOE 4305 (undergraduate) and BIOE 6305 (graduate), 3 credit hours. Taught every year since 2016.

We cover brain-machine interfacing with invasive and noninvasive recordings, including intracortical spikes, local field potentials, electrocorticography, depth electrode recordings and EEG. Students work with real neural datasets. They implement encoding and decoding models (population vector, linear, Kalman filter and neural network decoders) and evaluate closed-loop BMI systems. We also spend time on the translational and ethical questions. Reinforcement learning comes up again and again. At the end, each student does an independent project that replicates and extends a recent BMI or neural decoding study using publicly available data.

Introduction to Neural Computation / Advanced Artificial Neural Networks

BIOE 4313 (undergraduate) and BIOE 6306 (graduate), 3 credit hours. Taught every year since 2017.

We start with the math of how a network learns (gradient descent and backpropagation) and work up through the rest of modern deep learning, convolutional and recurrent networks first, then attention and transformers. From there we get into how large language models are pretrained, and after that how they're adapted with parameter-efficient fine-tuning and aligned with human feedback. We also cover generative models, including variational autoencoders, GANs and diffusion models. Students work in Python with fastai and PyTorch on free Google Colab resources. For the final project, each student replicates a research paper published within the past year on their own and then extends it.