Explain streaming k-means I need to cluster 22 million vectors, each of which has a dimension of 1024. Can anyone explain streaming k-means to me?
 A: "Streaming kmeans" typically refers to what is also known as "online kmeans". Online algorithms refer to computations that are performed iteratively, with data arriving during the computation in single observations or in batches, in contrast to "offline" algorithms, where all the data are available when the computation starts. Typically, some intermediate result is calculated based on initial data and then modified as new data arrive. For instance, exponential smoothing is a forecasting algorithm that is very naturally performed online.
In the specific case of k-means, we would first apply a standard k-means algorithm to cluster an initial dataset. Then, the cluster centers would be updated as new data arrive. Such algorithms often keep and update clusterings with different numbers of clusters, because the optimal number of clusters may change over time as data arrives.
Alternative names for online algorithms include "sequential algorithms", and a number of other synonyms as per Wikipedia. For the specific case of online k-means, this SO question looks like a good place to start: Online k-means clustering
