Pan: Drag the map with the mouse. Zoom: Press the ALT-key and turn the mouse-wheel.
This document is a hybrid of a map and commenting text
underneath which in turn points to references which back up the
connections.
In contrast to “self-organised” models which assume that
“objective” data organises itself to a brain map, this project
strongly believes that curation by a human is
essential. There are a number of reasons:
- Data and papers are biased: We assume every
paper and every dataset is biased – be it subconscious,
conscious or deliberate. It should be a no-brainer as it's generated
by humans for humans. It is thus important to compare different
views and then draw conclusions which most likely reflects what is
actually happening in the rat brain. I'm a computational
neuroscientist. This is both benefit and curse. The advantage is
that I can step back and can talk to different camps to find out
different intentions, slants or spin. On the other hand the danger
is that I force popular machine learning models onto
neuroscience. For that reason this map here makes no attempt to be
mathematically pure. For example there is no attempt to force TD
or deep learning onto this map.
- Uncertainty of the data: Neurophysiological data
rarely provides 100% certainty and a clear-cut
interpretation. Data is noisy and there will be almost every time a
study which contradicts the current state of the art. However,
through experience one learns to distinguish between spurious
results and those which will consolidate. This is not a simple
process eliminating outliers but to look at the studies from the
curator's experience because a spurious result can still trigger a
large amount of publications and again requires stepping back and
evaluating publications with possibly at a 10 year timeframe to check
which of these findings have been reproduced from different sources
and over years.
- Closed loop / Embodied: Crucially, an animal needs to
act in its environment because the limbic system makes an animal
finding rewards (and avoiding punishments). This means that the
animal generates actions and evaluates the resulting sensor inputs
which in turn generate actions and so forth. The whole map needs to
be able to establish a (real or simulated) rat performing reward
related behaviour, for example reversal learning. This constraint
demands that the limbic map needs to be seen always as a whole
embedded in its environment.
In summary: human curation of data is as important as humans
generating both data and publications. So far no algorithm can
fix this. A brain searching for rewards cannot do this while swimming
in a vat and thus requires to be embodied.
