Have a new PhD student starting and am unsure how to design their project on the "difficulty but ambitious" vs "easy but incremental" axis. What are your thoughts/strategies on this topic?

I feel like I have under-thought this aspect in the past, just picked what seemed most exciting scientifically and ramped up from there. 

 This video (showing a EM reconstruction of a 6um cube of rat hippocampus) completely changed how I think about the brain when I first saw it a few years ago.

It's all a jumbled mess of spaghetti, not a bunch of circles connected by lines as those pesky theorists would you believe.

https://youtu.be/Xhfnp2ZS0I8 

 In neurophysiology people often estimate "phase coherence", ie the trial-to-trial alignment of some brain signal (EEG/MEG/LFP) to some oscillation.

This is hard using frequentist stats, so Sydney Dimmock, me & @2ee6ec85  developed a better Bayesian method.

Now in press in eLife: https://elifesciences.org/articles/84602 

 Synaptic transmission is noisy, presumably because the brain tries to keep energy costs down.

When we modelled this energy cost in an ANN, it spent its energy cleverly: it made synapses that were important for the task more precise, and synapses that were less task-important more noisy.

Kind of shockingly, this is also what the brain should do if it were being Bayesian.

Top work by PhD student James Malkin, with @2ee6ec85 and @489525f9

https://arxiv.org/abs/2309.03194

https://static.mstdn.science/media_attachments/files/111/023/119/444/027/025/original/fcded3484283e637.jpeg