
Neuroscience postdoc @ucl. Using machine learning and computational modelling to help understand memory and navigation. Very amateur footballer for @Monstars_UK
United Kingdom
Joined September 2015
- Tweets426
- Following359
- Followers546
- Likes1.4K
Finally, hippocampus represents multiple spatial scales - particularly along the DV axis. What happens if there's plasticity between them? Well unless these multiple scales are segregated (e.g DV axis), the small SR place fields (short timescale SR's) get swallowed up by big ones
1
3
345
The resulting STDP-SR model place cells can explain a whole bunch of characteristics observed in real CA1 place fields, such as experience-dependent skews and shifts (note since our place fields are not anchored to a grid-world representation, we also see shifts in field peaks)
1
4
275
Even more cool, @TomNotGeorge simulated this for a whole range of parameters values, and found the ones derived from biological data were pretty much optimal! (with the exception of firing rate which obviously has biophysical & metabolic limitations)
1
4
289
We found that theta sweeps accelerate learning; reduce the noise in weights; and preserve learnt sequence structure capturing behavioural biases. Interestingly, switching off theta sweeps learns something more like an unbiased SR (keep in mind only ~50% place cells phase precess)
We learnt two other groups were working on different, but not exclusive, mechanisms to tackle the same problem! We coordinated submissions and incredibly these were all published alongside each other in @eLife yesterday! Big props to eLife for facilitating this triple submission
1
6
380
We thought theta phase precession, in particular theta sweeps, might achieve this by compressing behavioural sequences onto a timescale conducive to learning between spiking neurons (STDP), whilst preserving sequence structure
1
3
569
#cosyne2022 come check out the insanely talented @NotGeorgeTom present our poster I-050 8pm tonight (17th) on learning predictive representations using STDP and phase precession! Work done with the awesome @neuro_kim & @caswellcaswell
Hey #SfN18 ! Tomorrow 1-5pm I'll be presenting a poster on our work comparing rat, human and RL algorithm behaviour on a transition revaluation navigation task. Hope to see you there!











