Decoders

Question 1

Type of filtering for spikes in extracellular recording

Answer 1

High-pass filtering.

Question 2

Filtering for local field potentials (LPFs) in extracellular recording

Answer 2

Low-pass filtering.

Question 3

Extracellular recording

Answer 3

Can distinguish individual neurons 0-100μm from tip. Spikes detectable (but too noisy) grouped as multi-unit activity up to 150μm.

Question 4

3 steps of population analysis

Answer 4

1. Record from multiple neurons/sites (often simultaneous).
2. Spike sorting algorithms extract discrete spike times from continuous data.
3. Information inferred from spike trains using decoding algorithms (predict what is being perceived) or information theory (quantify info carried).

Question 5

Neural decoding

Answer 5

Analysing and categorising neuronal data to make a formal prediction of what (stimulus/behaviour) elicits a particular neural response.

Question 6

5 examples of decoding algorithms

Answer 6

1. Bayesian decoders (posterior probability).
2. Nearest neighbour decoders.
3. Fisher linear discriminant algorithms (LDA).
4. Support vector machines (SMV; high dimensional hyperplanes).
5. Artificial neural networks.

Question 7

How are decoders optimised?

Answer 7

Training set.

Question 8

How is decoder performance tested?

Answer 8

Cross-validation.

Question 9

What is a confusion matrix?

Answer 9

Shows the relative number of times a decoder predicts a stimulus/behaviour. Perfect decoding would have all entries on the diagonal.

Question 10

4 sources of information loss from extracting info from neuronal activities.

Answer 10

1. Binning.
2. Spike average counts (lose timing information).
3. Missing info about unlikely stimuli.
4. Using the wrong epoch window/not knowing the epoch window.

Question 11

Information theory

Answer 11

Quantifies total information the neural response contains (not just the most likey stimulus).

Question 12

2 issues with information theory

Answer 12

1. Over-estimate information (not all info used by brain).
2. Curse of dimensionality.

Question 13

Ezzyat et al. (2018) stimulation of MTL

Answer 13

Potentially improve memory (closed-loop stimulation therapies).

Question 14

Neural synchrony

Answer 14

Communication through coherence. Neurons that fire in synchrony facilitates inter-area neuronal communication.

Question 15

3 putative functions of phase coherence

Answer 15

1. Communication through coherence (neural synchrony).
2. Coincidence detection (synchrony aligns synaptic inputs).
3. Neural plasticity.

Question 16

3 types of multi-electrodes

Answer 16

1. Semichronic microdrives (more neurons over time).
2. Utah arrays (lots of electrodes per region).
3. Neuropixels probes.

Question 17

Granger causality )G-causality)

Answer 17

A statistical test for predictive causality. If information about the past of variable X helps predict the future of variable Y better than using only Y’s past information.

Question 18

How to improve akinesia PD?

Answer 18

Stimulate PPN (pedunculopontine nucleus).

Question 19

Brain-machine interfaces (BMI)

Answer 19

A system that can interface the brain with computers. Can ‘write-in signals’ through electrical stimulation (send info into the brain) or ‘read-out’ to decode intent.