When: Thursday 21st June 2018, 13:00-14:00.
Included in meeting:
The fortnightly BRAG meeting will be held tomorrow, 21/06/18 at 1pm in room Y801. This week we will have presentations by Laura and Victor.
Laura’s Talk: Developing a Bayesian Spatio-Temporal Model for Cyclist Exposure to Air Pollution: using detailed mobile sensor data to improve personal exposure estimates
Abstract : Current smart phone applications providing personal air quality exposure information and warnings base their air quality estimates on measurements from stationary monitoring stations or satellites, often only available for one specific city or on a large, city wide scale.
To obtain a more accurate and detailed personal air quality exposure estimate, for example for an individual cycling through Brisbane city centre, it is important to measure and model air quality at a much higher spatial and temporal resolution. To achieve this we have used a single air quality mobile sensor, the cutting edge in sensor design, mounted to a bicycle, to collect air quality data within Brisbane city centre. In this talk I will discuss how I developed the data collection design, my initial exploratory data analysis and how this will develop into a sophisticated, detailed model for air quality exposure for cyclists in Brisbane.
Victor’sTalk: Simple and robust algorithms for online learning
Abstract: In online learning, the data arrive sequentially. One can first design an algorithm for treating the data based on the assumption that the data is generated by a specific type of process. This works well as long as the assumption is verified but can fail otherwise. Alternatively one can avoid making any assumption and design algorithms that are robust to data that could be designed by a malignant adversary. However, this second option can lead to designing algorithms that are overly conservative and will not take advantage of cases where the data is easy to learn from. Can we have the cake and eat it? Or maybe eat 75% of the cake and still have 75% of the cake?
We illustrate this discussion in the problem of prediction with expert advice and the best-arm identification problem.
We look forward to seeing you all there!
Trish and Farzana