Justin Jude

BrainGate and MGH Center for Neurotechnology and NeuroRecovery

Providence, RI/ Boston, MA

I am currently a postdoctoral researcher at BrainGate(Brown University) and an appointed Research Fellow at Massachusetts General Hospital and Harvard Medical School. I am primarily concerned with decoding speech from neural data associated with chronically implanted multi-electrode arrays of human participants in the BrainGate2 clinical trial. I am also interested in improving the long term stability of real-time neural decoders of movement and speech.

Previously I was a PhD student in the Institute for Adaptive and Neural Computation at the University of Edinburgh where I worked with Matthias Hennig on latent variable modelling of neural data using sequential autoencoder networks.

I also looked at hippocampal representations in neural networks and how these form throughout training on maze tasks. We found that these formed representations in neural networks are strikingly similar to their neuronal counterparts. Additionally, I explored the potential and variety of biologically plausible neural networks, namely on training methods for Spiking Neural Networks and the applications and performance of such networks when compared with traditional neural networks.

Before starting a PhD I was a Machine Learning Engineer at PaperCup, working on Deep Generative Models such as WaveNet in the realm of Neural Speech Synthesis and Machine Translation.

I received my integrated Masters from University College London where I worked with Simone Severini and Edward Grant on Deep Generative Models such as Variational Autoencoders for Quantum Computing States. Here is a link to my masters thesis.

My undergraduate degree was also at University College London in Mathematics and Computer Science before focusing on Machine Learning, Deep Learning, Reinforcement Learning and Machine Vision during my integrated masters. Over the course of my undergraduate degree I squeezed in three internships at Santander, BAE Systems and Amazon.

Curriculum Vitae

selected publications

ICML

Robust alignment of cross-session recordings of neural population activity by behaviour via unsupervised domain adaptation

Jude, Justin, Perich, Matthew G, Miller, Lee E, and Hennig, Matthias H

International Conference on Machine Learning 2022

Abs HTML PDF

Neural population activity relating to behaviour is assumed to be inherently low-dimensional despite the observed high dimensionality of data recorded using multi-electrode arrays. Therefore, predicting behaviour from neural population recordings has been shown to be most effective when using latent variable models. Over time however, the activity of single neurons can drift, and different neurons will be recorded due to movement of implanted neural probes. This means that a decoder trained to predict behaviour on one day performs worse when tested on a different day. On the other hand, evidence suggests that the latent dynamics underlying behaviour may be stable even over months and years. Based on this idea, we introduce a model capable of inferring behaviourally relevant latent dynamics from previously unseen data recorded from the same animal, without any need for decoder recalibration. We show that unsupervised domain adaptation combined with a sequential variational autoencoder, trained on several sessions, can achieve good generalisation to unseen data and correctly predict behaviour where conventional methods fail. Our results further support the hypothesis that behaviour-related neural dynamics are low-dimensional and stable over time, and will enable more effective and flexible use of brain computer interface technologies.
NeurIPS

Capturing cross-session neural population variability through self-supervised identification of consistent neuron ensembles

Jude, Justin, Perich, Matthew G, Miller, Lee E, and Hennig, Matthias H

Proceedings of the 1st NeurIPS Workshop on Symmetry and Geometry in Neural Representations 2022

Abs HTML PDF

Decoding stimuli or behaviour from recorded neural activity is a common approach to interrogate brain function in research, and an essential part of brain-computer and brain-machine interfaces. Reliable decoding even from small neural populations is possible because high dimensional neural population activity typically occupies low dimensional manifolds that are discoverable with suitable latent variable models. Over time however, drifts in activity of individual neurons and instabilities in neural recording devices can be substantial, making stable decoding over days and weeks impractical. While this drift cannot be predicted on an individual neuron level, population level variations over consecutive recording sessions such as differing sets of neurons and varying permutations of consistent neurons in recorded data may be learnable when the underlying manifold is stable over time. Classification of consistent versus unfamiliar neurons across sessions and accounting for deviations in the order of consistent recording neurons in recording datasets over sessions of recordings may then maintain decoding performance. In this work we show that self-supervised training of a deep neural network can be used to compensate for this inter-session variability. As a result, a sequential autoencoding model can maintain state-of-the-art behaviour decoding performance for completely unseen recording sessions several days into the future. Our approach only requires a single recording session for training the model, and is a step towards reliable, recalibration-free brain computer interfaces.