There is an increasing trend towards combining different kinds of data in applications of AI. You are required to outline an initial design for a system, based on a deep neural network, for estimating the probability of rainfall at a specific location, 6 hours into the future. The system will be given two kinds of input for the current time: (1) a single satellite image (such as that below) showing cloud cover over the wider region surrounding the location, and (2) the temperature and wind direction at ten ground-stations distributed over the same region.
1) Describe the architecture of your system, including as appropriate: the precise representation of inputs and output, any component neural networks used, types and number of layers, numbers of feature maps, use of activation functions, and resizing operations. You are not expected to undertake any implementation.
2) Give details of how you would train and test the system, including the data you would need, how the training is carried out, choice of the loss function, and at least one performance measure.
3) Give one strength and one weakness of your chosen design. less
View job posting