Skip to main content Skip to navigation

Towards a functional model for associate learning and memory formation

Primary Supervisor: Dr Jian Liu, Department of Neuroscience, Psychology and Behaviour

Secondary supervisor: Rodrigo Quian Quiroga

PhD project title: Towards a functional model for associate learning and memory formation

University of Registration: University of Leicester

Project outline

The aim of this project is to study the most up-to-date experimental data regarding single-neuron and network learning and coding in animals and humans, with the expectation that a functional model could be established thereon. If possible, this model shall be not just biologically descriptive but also computationally implementable, as the stimulation protocol is based on the natural scenes of visual and auditory scenes that are beyond the simple protocols.

Associative learning is one of the most fundamental modes of learning and memory formation in animals and humans alike. However, the neuronal models underlying associative learning remain elusive. Up to now, two major lines of competing models are present, i.e., the parallel distributed processing (PDP) models and the sparse coding representation (SCR) models. PDP models state that all the items are learned and stored by the entire neuronal network with the synaptic weight matrix as an ensemble; whereas SCR models depict that each individual item is learned and stored by a selective sparse subset of neurons. While PDP models have been prevailing in machine learning technologies, SCR models are becoming greatly substantiated by recent advances in experimental neuroscience, in particular, human recordings conducted at the Centre for Systems Neuroscience at the University of Leicester.

This project involves data analysis where the data of animal and human recordings will be provided by our collaborators. We will have access to the data of single-cell hippocampal recordings during memory and learning tasks of human subjects under the natural stimulations, as well as the population of neurons recorded in awake animals. Based on these data, we will draw a hypothesised theoretical model that is plausible for these new sets of experimental data as well as for other related published results. Furthermore, designs of new experiments based on the theoretical predictions shall also be made such that the model could be experimentally testified or falsified.

A few pieces of descriptive experimental findings or theoretical expectations shall be compatible with the new model, including but not restricted to the following issues: (1) Time: the procedure of learning can be variable over multiple timescales, but the outcome of learning must be largely invariable over the objective observer’s timescale. (2) Space: the theoretical capacity of memory (reliable and specific representation of learned items) shall be predictable for any limited total number of neurons, such that it is experimentally testable. (3) Biological plausibility: the model must not contain any biologically implausible settings and operating rules, e.g.: a 1000-layer neural network, each of which is updated at microseconds timescale.

We will bring forward a new hypothesised model of associative learning and memory formation, refine and validate the model with theoretical calculations/simulations, justified the results by using experimental data provided by the valuable human single-cell recordings and the state-of-art animal recordings.

BBSRC Strategic Research Priority: Understanding the Rules of Life: Neuroscience and behaviou

    Techniques that will be undertaken during the project:

    • Computer programming based on Python, C/C++, MATLAB
    • Neural and neural network modelling
    • Statistical/Machine learning

    Contact: Dr Jian Liu, University of Leicester