The goal of this challenge is to determine the First Arrivals of earthquakes. Before formulating our specific problem, let us consider a short example of a similar problem which may help to understand the general idea.

We have a smartphone and we want to ask an audio assistant (e.g. "Ok Google!") to find some information on
the internet.

The audio assistant has to continuously record our voice from the microphone and when we speak the keywords
"Ok Google!", it activates speech recognition. To catch the keywords, the program inside of your smartphone
has to identify your voice among background noises and understand when your speech starts and ends. This
is called voice activity detection. 

Let us come back to our problem, where "voice" is earthquakes and "microphone" is geophones recording quakes (receiver for short). Instead of detecting full voice activity, we want to identify only when it starts. In seismology, this procedure is called first-arrival picking. In other words, we want to understand when an earthquake starts in time.

We will use Mean Absolute Error (MAE) to evaluate the performance of the submitted prediction results and rank the best-performed models.