Floating point Sparse Distributed Memory
GreyMemory is a new implementation of sparce distributed memory. Currently, we use GreyMemory as a time series anomaly detector.
Anton Mazhurin
Nawwaf Kharma
We implemented a set of completely new features:
- Floating point input data
- Extended storage enabling storing long time series
- We define and use resolution for input data, rather than a range of values. No more ranges for input channels.
- We use hyper-rectangle to allocate hard locations. In the original SDM an activation radius is used.
- We dynamically allocate hard locations on the fly
- We introduced a notion of META data. Meta data might be used for instance to retrieve an index of a long time series.
- We use a sliding window to predict store time series. It means greymemory is able to predict the next sample in the future on the fly.