Technical details about what we're doing:
We're working on a time-series problem called a Sequence Classification. Recurrent Neural Networks called Long Short-Term Memory (LSTM) networks are good for that. It’s what they use for natural language processing because it can remember context from the sentence you’ve just spoken. In this case its used to analyze not just the current timestamp of data, but what has been happening for the last 20 - 250 ms, which helps to reject false alarms.