Characterizing success pathways in networked graphs
View Patent ↗Methods and systems for inferring performance include a chain graph that has multiple nodes, one or more of which has a known value at a first timestamp. A value associated with a first node at the first timestamp influences a value associated with a second node connected to the first node by an edge at the first timestamp. Each node has an associated chain that represents values of the node at different timestamps. Known values of each node in the graph are propagated within each timestamp to all neighboring nodes. Known values of each node in the graph are propagated across different timestamps.
1. A method for inferring performance, comprising:
constructing a chain graph comprising a plurality of nodes, a connection value that characterizes a strength of an influence between pairs of nodes, and a temporal influence parameter that characterizes a strength of an influence between values for each node at different timestamps, one or more of said plurality of nodes having a known value at a first timestamp, wherein a value associated with a first node at the first timestamp influences a value associated with a second node connected to the first node by an edge at the first timestamp, and wherein each node has an associated chain that represents values of the node at different timestamps;
propagating known values of each node in the graph within each timestamp to all neighboring nodes using a processor; and
propagating known values of each timestamp in each node in the graph across different timestamps of each respective node in the graph using the processor.
2. The method of claim 1 , further comprising:
determining whether the chain graph has met a convergence criterion; and
repeating said steps of propagating if the chain graph has not met the convergence criterion.
3. The method of claim 1 , wherein a state of the chain graph at each timestamp is constructed as a respective matrix that includes each known value of each node in the chain graph along the matrix's diagonal and the connection values.
4. The method of claim 3 , wherein the chain graph is constructed as a block-tridiagonal matrix that includes a state matrix for the graph at each element of the diagonal at respective timestamps.
5. The method of claim 4 , wherein the block tridiagonal matrix includes the temporal influence parameter.
6. The method of claim 1 , wherein propagating known values across different timestamps is performed in parallel.
7. The method of claim 6 , wherein propagating known values across different timestamps is performed using pointer jumping.
8. The method of claim 7 , wherein pointer jumping is performed in log(T) steps, where T is a number of timestamps in the nodes' chains.
9. A computer readable storage medium comprising a computer readable program for inferring performance, wherein the computer readable program when executed on a computer causes the computer to perform the steps of claim 1 .
10. A method for inferring performance, comprising:
constructing a chain graph comprising a plurality of nodes, one or more of which has a known value at a first timestamp, wherein a value associated with a first node at the first timestamp influences a value associated with a second node connected to the first node by an edge at the first timestamp, and wherein each node has an associated chain that represents values of the node at different timestamps;
propagating known values of each node in the graph within each timestamp to all neighboring nodes;
propagating known values of each timestamp of each node in the graph across different timestamps of each respective node in the graph;
determining whether the chain graph has met a convergence criterion; and
repeating said steps of propagating if the chain graph has not met the convergence criterion,
wherein a state of the chain graph at each timestamp is constructed as a respective matrix that includes each known value of each node in the chain graph along the matrix's diagonal and connection values that characterize a strength of an influence between each pair of nodes at off-diagonal positions, and
wherein the chain graph is constructed as a block-tridiagonal matrix that includes a state matrix for the graph at each element of the diagonal at respective timestamps and further includes a temporal influence parameter that characterizes a strength of an influence between values for each node at different timestamps.
11. The method of claim 10 , wherein propagating known values across different timestamps is performed in parallel using pointer jumping.
12. The method of claim 11 , wherein pointer jumping is performed in log(T) steps, where T is a number of timestamps in the nodes' chains.
13. A system for inferring performance, comprising:
a chain graph module configured to construct a chain graph having a plurality of nodes, a connection value that characterizes a strength of an influence between pairs of nodes, and a temporal influence parameter that characterizes a strength of an influence between values for each node at different timestamps, one or more of said plurality of nodes having a known value at a first timestamp, wherein a value associated with a first node at the first timestamp influences a value associated with a second node connected to the first node by an edge at the first timestamp, and wherein each node has an associated chain that represents values of the node at different timestamps; and
a temporal propagation module comprising a processor configured to propagate known values of each node in the graph within each timestamp to all neighboring nodes and to propagate known values of each timestamp of each node in the graph across different timestamps of each respective node in the graph.
14. The system of claim 13 , wherein the temporal propagation module is further configured to determine whether the chain graph has met a convergence criterion and to repeat propagation if the chain graph has not met the convergence criterion.
15. The system of claim 13 , wherein a state of the chain graph at each timestamp is constructed as a respective matrix that includes each known value of each node in the chain graph along the matrix's diagonal and the connection values.
16. The system of claim 15 , wherein the chain graph is constructed as a block-tridiagonal matrix that includes a state matrix for the graph at each element of the diagonal at respective timestamps.
17. The system of claim 16 , wherein the block tridiagonal matrix includes the temporal influence parameter.
18. The system of claim 13 , wherein the temporal propagation module is further configured to perform propagation across different timestamps in parallel.
19. The system of claim 18 , wherein the temporal propagation module is further configured to propagate known values across different timestamps using pointer jumping.
20. The system of claim 19 , wherein pointer jumping is performed in log(T) steps, where T is a number of timestamps in the nodes' chains.