Resource trading strategies with risk selection in collaborative training market

The rapid development of edge computing and artificial intelligence has brought growing interest in collaborative training. While prior research has addressed technical aspects of resource allocation, less attention has been paid to the underlying economic mechanisms of resource trading. In this study, we examine how task publishers can effectively allocate budgets between computational and data resources during co-training. To address the uncertainty in data acquisition, we introduce Constant Proportion Portfolio Investment approach to assist in the construction of the payoff maximization problem with budget constraints. With the aid of economic tools, we design Swing Gradient Search Algorithm to obtain the optimal investment portfolio strategy, thereby addressing the coupling relationship between the quantities of resource acquisition. We also explore how market dynamics evolve in response to changes in supply and demand. To maintain dynamic market equilibrium, we develop two types of pricing algorithms, one based on stepped price adjustments for selected sellers, and another based on smoothed adjustments for all sellers. Simulation results demonstrate that the proposed strategies and algorithms offer acceptable performance in terms of algorithmic efficiency and strategic effectiveness, while also preserving fundamental economic principles and supporting stable market dynamics.