Primary supervisorMehdi Adibi
The aim of this project is to understand the computations underlying animals’ choice in dynamic and changing environments. The natural environment is multisensory, dynamic and changing, requiring animals to continually adapt and update their learned knowledge of statistical regularities in the environment that signal the presence of primary needs like water, food and mates. Yet, how the brain adapts and updates itself to the non-stationary and dynamic attributes of natural environments remains unexplored. We trained rats in a two-choice sensory categorisation task that the categorisation boundary switches between two values. choices were rewarded by diluted fruit juice. The rats followed the change in the boundary after 20-30 trials. However, the animal’s choices in a proportion of trials depends on parameters other than the current stimulus. These factors include the history of previous choices, the outcome of previous choices (correct vs incorrect categorisation of stimulus). The aim of this project is to quantify the behavioural states where the choices are governed by non-sensory components such as attentional lapses and bias, or sensory cue, and then provide a mechanistic model that explains the behaviour of the animals. Direct work with animals is not required, however, if interested, there will be a unique opportunity to observe or contribute in animal experiments.
Matlab or Python, basic knowledge of optimisation and statistical models such as Hidden Markov Model and reinforcement learning is referred.