The fortnightly BRAG meeting will be held this Thursday 12th June at 1pm via Zoom/GP-Y801. This week we will have presentations by Sean Martin and Dr Ethan Goan.
Join Zoom Meeting
https://qut.zoom.us/j/83678271498?pwd=lpnfftGkFIgbskbcUhiUvHHvj0Ti6N.1&from=addon
Meeting ID: 836 7827 1498 / Passcode: 346383
Sean Martin’s Talk
Abstract: Islands are hotspots for biodiversity and biodiversity loss. Invasive species are a key driver of biodiversity loss on islands, so eradication of invasive species can be an effective conservation strategy. Eradication from islands is often feasible and effective, but knowing when to stop looking for an invasive population (ie when to declare success) can be difficult. Statistical models have been developed to infer the success of an eradication campaign based on historical detections of the target population, however none of these models incorporate variation into the detectability of individuals that may arise from different microhabitats or behaviors. Deriving a closed-form likelihood equation that incorporates variation and is tractable has proven difficult, so we are now approaching the problem with approximate Bayesian computation methods. However, eradication campaign managers rarely apply these statistical models due to various limitations in resources, skill, and realism. To develop a decision support tool that may find application, we also aim to derive a robust, easily applied heuristic from the analysis.
Dr Ethan Goan’s Talk
Uncertainty Quantification for Outlier Detection in Computer Vision
Abstract: For a well-calibrated model, we would like to exhibit greater uncertainty for out-of-distribution samples. This could occur naturally when applied to unfamiliar scenarios, but could also occur during the presence of intentional adversarial attacks. This talk will discuss how this uncertainty information could be used as a mechanism to detect these instances when applied to computer vision scenarios.