Dynamic Task Assignment Framework for Mobile Crowdsensing with Deep Reinforcement Learning

Table 1

Experiment definition parameters.

Parameters	Settings
RM (capacity of the replay memory)	50000
(probability of random selection)	0.2-0.1
(steps to update the target network)	300
(discount factor)	0.95
(learning rate)	0.001
Minibatch size	64
Sliding window	50
(maximum training episodes in Algorithm 1)	50000
(the maximum number of a task completed)	1-3
Periods in Gowalla	24
Periods in Foursquare	40
Number of workers in Gowalla	[60,70,80,90,100]
Number of workers in Foursquare	[30,40,50,60,70]
Task test sets days in Gowalla	[7,9,10,12]
Task test sets days in Foursquare	[10,12,15,20]
(perception radius)	[1 km, 2 km, 3 km]