- Synaptic compensation using remote memories (which of course are easier to obtain in AD) actually accelerates cognitive decline due to the decreased variability in the data set used to calculate compensatory terms.
- Where small-world connectivity occurs in the brain, this has the effect of increasing redundancy and resilience to damage, at the expense of lower overall capacity.
- Selectively partially muting, rather than deleting, synapses in a spreading area of damage can be used as a simulation of tau pathology, in which vesicles become blocked and axons degrade. This type of lesioning offers a much more graceful decline in performance as the compensatory mechanisms keep up with the changes, but catastrophic damage occurs after a certain level of lesioning and the decline in performance is much more dramatic than with plain synaptic deletion.
Once I get feedback from IJCNN and make any required changes, I'll put a copy of the paper up on my website. In the meantime, I need to think about next steps.
Firstly, my RSMG4 progress report is due in April. This will be a simple 2000-4000 word write-up of my progress over the last 6 months, including what I learned in Göttingen and Zürich on the neuroscience and reservoir computing courses, and of course the IJCNN paper.
Beyond that, I guess I will have the following tasks to choose from:
- [OPTIONAL] Continue work on the Ruppin and Reggia model -- design experiments to explore the above effects further.
- [ESSENTIAL] Collate definitive medical data against which my models should be compared, and lay out the way in which my model can be shown to be a small part of the overall larger brain organisation (i.e. hippocampal vs neocortical organisation). This is hard and will require much thought!
- Incorporate amyloid (including N-APP and anaesthesia) pathology simulations to test cutting-edge medical theories in computational networks.
- Begin working on implementation of a reservoir computing network which incorporates synaptic compensation (very important, as the reservoir network's dynamics change dramatically with only slight changes in the internal reservoir).
- Can reservoir networks be shown to be better models than, or at least as accurate as, Hopfield-style associative networks (a la Ruppin and Reggia)? What are the differences in behaviour?
- What other symptoms of AD can be represented in a reservoir network, other than just failure to accurately recall a stored pattern? As the computational power is much greater, could a basic model of degradation of language / motor skills or some other feature be implemented?
Lots to do!