Our next HTM Hackers’ Hangout will be a live Q/A session about anomaly detection. But I know some of you cannot make it live, so please leave your questions here. I’ve made a list of topics / questions that have come up in this subject in the past (below). If you have any others, please add them in your reply.
- How is the raw anomaly score calculated from the cellular state in HTM?
- Why does the raw anomaly score usually need a post process like “anomaly likelihood”?
- Does anomaly detection work on all the input fields in a model?
- What does the anomaly likelihood process actually do?
- Why does the anomaly likelihood threshold need to be so high? (0.9999)