Joint learning under incomplete information and dynamic MU availability

Design an efficient and scalable solution that jointly addresses incomplete information at the mobile crowdsensing platform and at mobile units while accounting for dynamic mobile-unit availability in mobile crowdsensing-based sensing-as-a-service systems.

Background

The paper formulates mobile crowdsensing as a dynamic two-sided matching game between mobile units (MUs), which propose sensing tasks, and the mobile crowdsensing platform (MCSP), which assigns tasks. Unlike conventional matching models, MU availability changes over time because battery-powered devices have limited and dynamically replenished energy. The task proposal and assignment decisions are also coupled, while sensing quality, task effort, preferences, and future availability are not known in advance.

The cited challenge concerns developing a joint learning and orchestration mechanism that remains efficient and scalable despite incomplete information on both sides of the market and endogenous, energy-dependent MU availability. The paper proposes Energy-aware Two-Sided Learning (ETSL) as a solution to this challenge, so the problem is stated as an open problem motivating the work rather than as an unresolved issue after the paper's proposed method.

References

Although previous works have contributed significantly to task proposal and task assignment in \gls{mcs}, designing an efficient and scalable solution that jointly addresses incomplete information at the \gls{mcsp} and at the \glspl{mu}, while accounting for dynamic availability, remains an open problem.

— Energy-Aware Two-Sided Learning for Dynamic Matching Games in Mobile Crowdsensing  (2609.29880 - Dongare et al., 24 Sep 2026) in Section 1, Introduction

Remark: Propositions~\ref{prop:etp_convergence} and~\ref{prop:tslta_convergence} characterize the fixed point each side converges to under stationarity of the other, because both sides learn simultaneously in practice, this is only approximately satisfied at finite $t$. The two updates operate on different effective timescales ($\tilde{U}{\mathrm{MCSP}_{k,z,t}$ updates once per assignment, while $Q(S_{k,t}, A_{k,t})$ updates once per MU decision), which is consistent with the two-timescale stochastic approximation framework of; we leave a full joint convergence proof to future work and instead provide extensive empirical evidence of stability in Sec.~V.

— Energy-Aware Two-Sided Learning for Dynamic Matching Games in Mobile Crowdsensing  (2609.29880 - Dongare et al., 24 Sep 2026) in Section 4, subsection “Convergence of ETSL,” remark following Proposition 2