IPAB Seminar - 08/06/2023

Speaker: Evripidis Gkanias

Title: Memory dynamics in the fruit fly brain

Abstract: Many proposed neural circuits for learning are theoretically well-grounded but have limited biological evidence to support them. Emerging anatomical and functional data from neurons in the fruit fly brain allows us to identify interconnections in the mushroom body neuropil that form an ‘incentive circuit’, supporting rich dynamics of memory acquisition, consolidation and forgetting. In addition, new insights into dopamine function in the fruit fly brain motivate a novel ‘dopaminergic plasticity rule’ that alters the weights between input sensory patterns and output neurons of the mushroom body. Connections between output and dopaminergic neurons allow the flexible formation of short-term memories, transfer to long-term memory, and balancing between opposing valences for similar stimuli, resulting in an effective exploration-exploitation trade-off. When this circuit was used for controlling the behaviour of simulated flies, it replicated many features of fly behaviour such as the differential acquisition of aversive and attractive memories despite symmetry in the neural circuitry. This provides new insight into the underlying mechanisms of associative learning that might generalise beyond the fruit fly brain.

DOI: https://doi.org/10.7554/eLife.75611

Speaker: Craig Innes

Title:  Automated Vehicle Failure Probabilities for Highway Traffic Rules using Perception Error Models

Abstract: Assessing the probability that an automated vehicle (AV) does not violate complex traffic rules in simulation poses two main problems: First, how do we assess the danger of real-world sensor failures (e.g., from cameras and lidar) in a simulated environment? Second, if violations of safety properties are extremely rare, how can we get low-variance, reliable estimates of the failure probability given an efficient number of simulations?

This talk will deal with both issues using a combination of Perception Error Models and Adaptive Importance Sampling.