Distributed approach for feature modeling using principal component analysis
In one embodiment, techniques are shown and described relating to a distributed approach for feature modeling on an LLN using principal component analysis. In one specific embodiment, a computer network has a plurality of nodes and a router. The router is configured to select one or more nodes of the plurality of nodes that will collaborate with the router for collectively computing a model of respective features for input to a Principal Component Analysis (PCA) model. In addition, the selected one or more nodes and the router are configured to perform a distributed computation of a PCA model between the router and the selected one or more nodes.
1. A system, comprising:
a plurality of nodes in a computer network; and
a router in the computer network, the router configured to select one or more nodes of the plurality of nodes that will collaborate with the router for collectively computing a model of respective features for a Principal Component Analysis (PCA) model,
wherein the selected one or more nodes and the router are configured to perform a distributed computation of a PCA model between the router and the selected one or more nodes, and
wherein the router is configured to generate a list of candidate nodes for the selected one or more nodes, the list containing nodes of the plurality of nodes whose features are to be modeled.
2. The system as in claim 1 , wherein the router is further configured to limit the list of candidate nodes based on characteristics selected from a group consisting of: location, type, and property.
3. The system as in claim 1 , wherein the router is further configured to filter the list of candidate nodes to those nodes that can effectively participate in the model computation.
4. The system as in claim 1 , wherein the plurality of nodes are configured to refuse selection.
5. The system as in claim 1 , wherein the selected one or more nodes are configured to perform the distributed computation by locally collecting features, locally computing a summarization of the collected features, and communicating the summarization with the router.
6. The system as in claim 5 , wherein the summarization comprises a mean of the collected features.
7. The system as in claim 5 , wherein the router is further configured to compute a global summarization of the collected features, and communicate the global summarization with the selected one or more nodes.
8. The system as in claim 7 , wherein the router is further configured to communicate the global summarization with a central computational entity.
9. The system as in claim 7 , wherein the router is further configured to perform an orthogonalization of the collected features.
10. The system as in claim 1 , wherein the router and selected one or more nodes are further configured to update the distributed computation.
11. The system as in claim 1 , wherein the router is further configured to explicitly stop the distributed computation based on one or both of time and number of collected features.
12. The system as in claim 1 , wherein the selected one or more nodes are further configured to aggregate summarizations of other selected nodes as the summarizations of other selected nodes pass through the selected one or more nodes toward the router.
13. A method, comprising:
selecting, by a router, one or more nodes of a plurality of nodes in a computer network that will collaborate with the router for collectively computing a model of respective features for input to a Principal Component Analysis (PCA) model;
performing a distributed computation of a PCA model between the router and the selected one or more nodes; and
generating a list of candidate nodes for the selected one or more nodes, the list containing nodes of the plurality of nodes whose features are to be modeled.
14. The method as in claim 13 , further comprising:
limiting the list of candidate nodes based on characteristics selected from a group consisting of: location, type, and property; and
filtering the list of candidate nodes to those nodes that can effectively participate in the model computation.
15. The method as in claim 13 , further comprising:
computing a global summarization of collected features from the selected one or more nodes; and
communicating the global summarization with the selected one or more nodes.
16. The method as in claim 15 , further comprising:
performing an orthogonalization of the collected features.
17. The method as in claim 13 , further comprising:
updating the distributed computation.
18. The method as in claim 13 , further comprising:
explicitly stopping the distributed computation based on one or both of time and number of collected features.
19. A method, comprising:
determining selection of a particular selected node of one or more nodes that will collaborate with a router for collectively computing a model of respective features for input to a Principal Component Analysis (PCA) model; and
performing a distributed computation of a PCA model between the router and the selected one or more nodes, wherein performing the distributed computation comprises:
locally collecting features,
locally computing a summarization of the collected features, and
communicating the summarization with the router.
20. The method as in claim 19 , further comprising:
refusing the selection.
21. The method as in claim 19 , further comprising:
updating the distributed computation.
22. The method as in claim 19 , further comprising:
aggregating summarizations of other selected nodes as the summarizations of other selected nodes pass through the selected node toward the router.