Memory consolidation and forgetting in natural and artificial neural circuits
Dr A Lin
Prof Eleni Vasilaki
No more applications being accepted
Competition Funded PhD Project (European/UK Students Only)
How should intelligent systems store learned information? Biological brains have evolved to retain important information as consolidated memories, while forgetting unimportant or obsolete memories. In contrast, artificial neural networks often undergo so-called ‘catastrophic forgetting’ when learning new information rather than selectively erasing irrelevant memories. Solutions to this engineering problem may find inspiration in how biological brains consolidate and forget memories.
Fruit flies learn to associate specific odours with rewards or punishments, but these memories fade over time, like human memories. Recent work indicates that forgetting in flies is not simply passive decay; rather, both forgetting and consolidation are active processes requiring persistent spontaneous neural activity in memory-encoding neurons after learning.
We will test whether such spontaneous activity can provide the computational basis for regulating memory consolidation and forgetting. First, we will experimentally measure this post-learning spontaneous activity and investigate the circuit mechanisms that regulate it. Second, we will computationally model alternative strategies for post-learning consolidation and forgetting in the fly brain, some derived from our experimental results, to test which, if any, are best able to flexibly retain and forget learned associations. Our results may suggest new strategies for flexible learning for artificial intelligence.
RCUK equivalent home stipend rate per annum for 3.5 years
Home tuition fees for 3.5 years*
£6500 for consumables
A first class or upper second class honours degree in a biological sciences subject or a related discipline, or a merit or distinction in a suitable MSc. We welcome applications from candidates from a range of backgrounds (from biology to computer science or physics), especially those with strong quantitative backgrounds.