2 Introduction to the book
2.1 What is computational neuroscience?
Computational neuroscience is just using computers to study brains.
What counts as brains has come to refer to pretty much anything across a very wide range of levels. E.g.,
Biochemical and electrical signal transduction and prorogation within individual cell compartments.
Intracellular signalling cascades.
Whole cell action potential spiking properties.
Collection of neurons as dynamical systems.
Local field potential, hemodynamic response, cortical oscillatory patterns.
Behaviour (e.g., decision making, action selection and execution).
Using computers to study purely neural systems – without concern for behaviour – is pretty fairly called computational neuroscience.
Using computers to study behaviour – without concern for neural systems – would fall more naturally under the banner of cognitive modelling.
Using computers to study how the brain drives behaviour – i.e., explicitly caring about both neural and cognitive domains and trying to link them – is sometimes given special names like computational cognitive neuroscience.
2.2 Marr’s levels
Computational: A specification of what a system does and why it does it.
Algorithmic: A specification of how it does what it does (i.e., what process is followed).
implementational: A specification of how the system is implemented in the brain.
In my opinion, few models cleanly reside at a single one of these levels and many probably are best classified between levels. Even so, the conceptual divisions they get us thinking about are useful to have on hand.
2.3 Why computers?
Mathematics is at the heart of computational neuroscience.
The goal is to write down equations that formally describe the process through which the phenomena of interest are hypothesised to emerge.
The need for computers arises when the mathematics that you write down is complicated and difficult to evaluate.
Computational approaches give us tools to deal with these difficult mathematical situations. Well, some of them at least.
2.4 What is the real value gained?
Have you ever stuggled or argued with yourelf or others about what a certain hypothesis predicts in a particular experiment?
Mathematical precision makes science happy. It eliminates ambiguity and thereby makes mathematical models more falsifiable.
In practice, the word more in the last sentence above can do an awful lot of work. Coming up with experiments that are strong tests of any model, even mathematically precise models, can be very difficult.
Even so, there is no question that mathematics and computers help… and sometimes they help a whole lot.
2.5 Why Python
- It is a real programming language
- It is mature
- It is widely embraced in psychology and neuroscience
- It is widely embraced inside and outside academia
- It is widely used in the machine learning community
- It is relatively easy to learn
2.6 Getting started with Python
I recommend getting started with Google Colab.
If you dislike the web notebook interface, then you might try Anaconda.
Anaconda will provide the IDE Spyder which isn’t a bad place to work with Python.
Ultimately, there are many Python programming tools to choose from. Use what you like.
2.7 Learning Python
Just like with other programming languages you have been exposed to, the base functionality of Python is extended by external libraries. We will make heavy use of the following libraries:
-
-
This one isn’t as beginner friendly. It’s just here for those of you with more programming experience that want to advance your skills and understand the language at a deeper level.
-
This will be the focus of week 2
-
This will be the focus of weeks 2 and 3