Sign in

James McAllister

@jajmca.bsky.social
105 followers 155 following 26 posts

Postdoc at Trinity College Dublin and Oxford University. Computational and Theoretical Neuroscience.

PostsRepliesMedia
James McAllister @jajmca.bsky.social · 29/06/2026
Just back from AREADNE, a conference on the Encoding and Decoding of Neural Ensembles, in Milos, Greece. It was an incredible week of sea, sun, and science. It emphasised to me again that science isn't just the methods, analyses, or results. It's the sharing, talking, listening, and being inspired.
282
James McAllister @jajmca.bsky.social · 16/06/2026
Appreciation post for my incredible supervisor Cian O'Donnell @cianodonnell.bsky.social. Such an inspiration & wonderful mentor as Supervisor & School Research Director. He organised a brilliant ISRC Discovery Day today @ulsteruni.bsky.social, showcasing the breadth & depth of our research.
161
James McAllister @jajmca.bsky.social · 05/06/2026
Beautiful beaches of Donegal | Tránna áille Dhún na nGall Dooey Carrickfinn Silver Strand Murvagh
040
James McAllister @jajmca.bsky.social · 02/04/2026
𝗧𝗵𝗲 𝘁𝗿𝗮𝗱𝗲-𝗼𝗳𝗳: The efficiency & robustness comes with a price. We found biological wiring leads to lower-dimensional, less expressive dynamics.
Lyapunov dimension of dynamics
140
James McAllister @jajmca.bsky.social · 02/04/2026
We find that neuron self-recurrency is a key feature which stabilises spectral properties & dynamics, inducing robustness to node loss & hyperparameter variation.
120
James McAllister @jajmca.bsky.social · 02/04/2026
𝗥𝗼𝗯𝘂𝘀𝘁𝗻𝗲𝘀𝘀: CoNNs also proved to be significantly more resilient to: - node loss, maintaining performance & spectral radius under pruning longer than conventional random networks - parameter perturbation, where CoNNs better maintained critical dynamics & smoother dynamical transitions.
Relative spectral radius under random pruning
120
James McAllister @jajmca.bsky.social · 02/04/2026
𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: We found that CoNNs generally achieved comparable task performance to classic Echo State Networks, but with lower “wiring costs” (smaller weights) & lower “energy demands” (reduced neural activity). CoNNs also showed more specialised engagement (smaller subsets of contributing neurons).
Participation ratios of neural engagement distributions
130
James McAllister @jajmca.bsky.social · 02/04/2026
Really excited to share a new preprint! 𝗛𝗼𝘄 𝗱𝗼 𝗯𝗿𝗮𝗶𝗻𝘀 𝘀𝘁𝗮𝘆 𝗿𝗼𝗯𝘂𝘀𝘁 & 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝘄𝗵𝗶𝗹𝗲 𝗯𝗲𝗶𝗻𝗴 𝗶𝗻𝗰𝗿𝗲𝗱𝗶𝗯𝗹𝘆 𝘀𝗽𝗮𝗿𝘀𝗲? We explore how! www.biorxiv.org/content/10.6... We built Connectome-based Neural Networks (CoNNs) using Drosophila wiring (larva&adult) & compared with random networks with same sparsity.
Constructing connectome-based neural networks
15213