Week 3.10 - Neural Decoding: Signal Reconstruction Flashcards
What is the definition of a feature?
Features are attributes of things in the world.
Features are important because they represent what is important about the data.
Name a few examples of features
For example
shape, object, orientation, edge, colour, smell, distance, rhythm, brightness, texture, temperature, amplitude, depth…
What is information entropy?
The average amount of information you get from one sample drawn from a given probability of a boolean variable.
Information entropy is usually measured in bits (0,1).
What is a bit?
A bit is short for binary unit, and is the smallest unit in a computer. Bits have a binary value of either 0 or 1.
One bit is the information you gain when you answer a yes and no question where the chances of each answer are 50-50.
What is a feature space?
Dimensions of some object that measures its characteristics. The number of dimensions is often equal to the number of measurable variables. Complex features can be combinations of lower level features.
What is the role of the Spike Triggered Average?
To discover what is the relationship between stimuli and spikes in a given neuron.
How do you use the Spike Triggered Average to reconstruct stimulus?
For a spike train from an unknown stimulus convolve the spike train with the spike triggered average.
What is information?
The classic definition of information is “a difference that makes a difference” which is a bit abstract. More concretely, information can be thought of a set of messages that are transmitted over a noisy channel.
Information theory deals with the question of what it takes to get information from one place to another, i.e. transmission. Information is anything that leaves you knowing more than you did before you got the information, usually referred to as reducing your uncertainty. An important measure in information theory is entropy.
Define what is ‘decoding’ in the context of spiking neural networks.
Decoding is to retrieve information from a spike train.
The general assumption is that neurons do not fire at random, there is structure in their activity.
The fact that there is structure in the firing patterns of neurons means that they must carry information. That is, the brain represents information with neurons and their spike trains.
Using specific decoding algorithms, we can take these rasters of spiking activity and output the predicted response. For example, when a person thinks about moving her arm, the decoding algorithm would translate that thought from the spiking activity.
In short, neural decoding attempts to understand how spikes elicit activity and responses in the brain; reading out the brain.
Define what is ‘encoding’ in the context of spiking neural networks.
Next to reading from the brain, we can also go the other way around. Neural encoding studies how input is transformed into spikes.
