This paper proposes novel approaches for jointly designing user equipment (UE) association and power control (PC) in a downlink user-centric cell-free massive multiple-input multiple-output (CFmMIMO) network. In this network, each UE is served by a set of access points (APs) to reduce fronthaul signaling and computational complexity. The goal is to maximize the sum spectral efficiency (SE) of the UEs. A mixed-integer nonconvex optimization problem is formulated under constraints on per-AP transmit power, quality-of-service rate requirements, maximum fronthaul signaling load, and maximum number of UEs served by each AP. To efficiently solve this problem, two different schemes are proposed based on the size of the CFmMIMO system. For small-scale systems, a successive convex approximation (SCA) method is used to obtain a stationary solution, and a learning-based method (JointCFNet) is developed to reduce computational complexity. For large-scale systems, a low-complexity suboptimal algorithm using accelerated projected gradient (APG) techniques is proposed. Numerical results show that JointCFNet achieves similar performance to the SCA algorithm in small-scale systems but with significantly reduced run time. The APG approach is confirmed to run much faster than the SCA algorithm in large-scale systems while achieving SE performance close to that of the SCA approach. Additionally, the median sum SE of the APG method is up to about 2.8 times higher than that of a heuristic baseline scheme. The paper also discusses the system model, problem formulation, and the proposed solutions for small-scale and large-scale CFmMIMO systems.This paper proposes novel approaches for jointly designing user equipment (UE) association and power control (PC) in a downlink user-centric cell-free massive multiple-input multiple-output (CFmMIMO) network. In this network, each UE is served by a set of access points (APs) to reduce fronthaul signaling and computational complexity. The goal is to maximize the sum spectral efficiency (SE) of the UEs. A mixed-integer nonconvex optimization problem is formulated under constraints on per-AP transmit power, quality-of-service rate requirements, maximum fronthaul signaling load, and maximum number of UEs served by each AP. To efficiently solve this problem, two different schemes are proposed based on the size of the CFmMIMO system. For small-scale systems, a successive convex approximation (SCA) method is used to obtain a stationary solution, and a learning-based method (JointCFNet) is developed to reduce computational complexity. For large-scale systems, a low-complexity suboptimal algorithm using accelerated projected gradient (APG) techniques is proposed. Numerical results show that JointCFNet achieves similar performance to the SCA algorithm in small-scale systems but with significantly reduced run time. The APG approach is confirmed to run much faster than the SCA algorithm in large-scale systems while achieving SE performance close to that of the SCA approach. Additionally, the median sum SE of the APG method is up to about 2.8 times higher than that of a heuristic baseline scheme. The paper also discusses the system model, problem formulation, and the proposed solutions for small-scale and large-scale CFmMIMO systems.