www-ai.cs.tu-dortmund.de/de/PERSONAL/MORIK/papers/pdf/Communication_efficient_learning_of_traffic_flow_in_a_network_of_wireless_presence_sensors.pdf
counts sent to neighboring nodes for learning
...
...
n1(j)
n2(j)
nc(j)
j
j j
D(j) j
x1(j)
x2(j)
x3(j)
y(j)1 . . . p 1 p
q1,low = 1
q1,high = 2
. . .
Fig. 1: Distributed Learning of Local Models
Let each node [...] nodes i = 1, . . . ,m. Each node i delivers an infi- nite series of real-valued raw measurements . . . , vt−1(i), vt(i), vt+1(i), . . ., where t denotes the current time step t and t− 1/t+ 1 denote next [...] follows:
For j = 1 to m do /* in parallel */
divide D(j) into batches B1, . . . , Bh
calculate label counts for each batch and store them in Q(j)
send Q(j) to nodes n1(j), . . . , nc(j)
For k = j, n1(j), . . …
