I really recommend sending your students to the Paris Spring School, many imaging and ephys techniques are covered in pretty good depth and the atmosphere is consistently fantastic every year! And in such a setting! parisneuro.ovh

As said earlier, this project was *very* collaborative. It has been an honour and a very fortunate learning experience to have the opportunity to work with many great scientists. Many thanks to the team led by Javier Medina, Dana Cohen, Steve Lisberger, Dimitar Kostadinov and Court Hull. 14/16

The VAE-compressed representations were then used to train a supervised deep neural network to predict neuron types directly (a multi-layer perceptron, or MLP). Our final classifier yields > 95% accuracy in both mouse and monkey cerebellar recordings. 12/16

First, we pre-trained two self-supervised ‘compressor networks’ with unidentified neurons (variational auto-encoders, or VAE). These networks learned to compress the electrical signature of neurons (their ‘waveform’ shape and temporal statistics) into a low-dimension space. 11/16

The second step was to map the electrical fingerprint of neurons onto their type. We developed a deep neural network for this task. A key to our success was using a hybrid self-supervised / supervised architecture, to end up with a semi-supervised classifier. 10/16

To answer this question, the first step was the creation of a carefully validated “ground truth” library of over 200 identified neurons. 8/16

How well the electrical fingerprints of neurons correlate with their type wasn’t clear. After all, the shape and frequency of voltage deflections on an electrode carry much less information than, say, the RNA profile or morphology of a neuron. 6/16

The power to ‘see’ every neuron type on your electrode comes with a challenge: it is hard to distinguish them! However, we are in luck: some neuron types fire big ‘spikes’ (voltage deflections), small spikes, fast spikes… Each neuron has a specific ‘electrical fingerprint’. 5/16

This work addresses a long-standing issue in neuroscience: the identification of neurons in extracellular recordings in vivo. 2/16

I am very proud to announce that my PhD paper finally came out in Cell! In this *very* collaborative study, we develop and release a deep-learning approach to predict neuron type identity from their electrical signature. doi.org/10.1016/j.ce... 1/16 🧵