Convolution operations
Convolution operations are key components of a CNN; these operations use an input tensor and a filter to compute the output. The key is deciding the parameters available to tune them.
Suppose we are tracking the location of an object. Its output is a single x(t), which is the position of the object at time t. Both x and t are real-valued, that is, we can get a different reading at any instant in time. Suppose that our measurement is noisy. To obtain a less noisy estimate of the object's position, we would like to average together measurements. More recent measurements are more relevant for us; we want this to be a weighted average giving higher weight to recent measurements. We can compute this using a weighting function w(a), where a is the age of a measurement (when the measurement was taken) If we apply a weighted average operation at every moment, we obtain a new function providing a smoothed estimate of the position of the object:
This operation is called convolution...
