Summary

Studied the literature in mammalian visual processing.
Studied methods, from information theory, that are able quantify natural information processing.
Explored a dataset containing Neuropixels (local field potential) recordings of neural activity in the primary visual cortex, taken from mice.
Came up with statistical experiments under an information-theoretic approach to investigate whether models of information transfer followed expectations from the neuroscience. We suspected there might be discrepancies, since neuroscience traditionally builds models of functional connectivity/effective connectivity using correlation in activity, rather than quantified information.
Found that information transfers in mice visual cortices either followed the expectations of hierarchical processing or were surprisingly homogenous.
Recommended future studies in task-related datasets to ascertain how much information relevant to a task is being processed in a recorded brain area.
Recommended future work in building information-theoretic estimators tailored to neuronal population dynamics.

Skills gained

Scientific writing.
Technical documentation.
Extensive literature research.
Inter-disciplinary communication (worked with neuroscientists, physicists and computer scientists).
Neuroscientific research.
Time-series analysis.
Neural signal processing (LFP recordings and spike trains).
Applications of
- information theory.
- complex systems modelling.
Statistical inference in an information theoretic/dynamical systems context.
- This is readily transferable to signal detection.
How to design an empirical model of a fundamental physical principle occurring in an observed system (in this case, that fundamental principle was information transfer).

Longer description

My honours supervisor (A/Prof Joseph Lizier) has been studying how information is processed in natural systems like brains. He takes a unique information-theoretic approach (after Shannon entropy etc.) that allows us to observe brain activity and then estimate the amount of information that has been processed.

My project took a specific look at information transfer (one of the stages of information processing) inside the visual cortices of mice. It was the first time an information-theoretic estimator of transfer in continuous-time was used on data from living creatures at this scale.

Broadly speaking the goal of the project was to investigate whether information flowed between regions in directions that one expects from the traditional neuroscientific analysis. The traditional approach is to study structural connections, or better yet to study bivariate or multivariate correlations in activity. However the approach followed by Professor Lizier and the growing numbers aware of information theory, allows a model of the amount of information present in the system, rather than just a model of correlated activity (what is usually called “functional” or “effective connectivity”).

Quantification of information processing can help address large questions in neuroscience like “how does a particular neuronal population encode information relevant to the task being performed?”

Briefly, the project found information flows that either matched expectations inside the visual cortex or were unexpectedly homogenous between cortical layers. This points to the versatility of how information can be represented and processed across structural connections that are relatively static. We suggested that future studies in neural dynamics during tasks and behaviours can adopt our information-theoretic approach to assess when, where and how much information relevant to a task is being processed inside a recorded brain area.

We also suggested that future studies can look at building information-theoretic estimators that rely on more emergent structures in dynamical systems; like bursts and oscillations. This would quantify information processing at various analytical levels of the dynamical system, not only of spike times of neuronal processes. Work with this flavour is being done at the University of Sydney, as a collaboration between three groups (including Joe’s).

I strongly recommend checking them out.

Submitted thesis

Code

Repository: spikes-information-transfer.