An Internet-of-Things (IoT) system typically comprises many small computing elements (nodes) that are battery-powered and communicate over a wireless network. These elements monitor properties in the environment and send the data to client applications via gateways. The wireless networks used by the elements are subject to uncertainties that are difficult to predict upfront, such as dynamic objects (swaying trees, cars, …) and changing weather conditions that may deteriorate the transmissions. To ensure reliable communication over a wireless network of energy-constrained elements, recent research has proposed self-adaptive IoT systems. Such a self-adaptive system equips the network of elements – referred to as the managed system – with a feedback loop – the managing system. The managing system monitors the changing conditions and adapts the transmission settings of the IoT network to ensure the system’s quality goals. Leveraging and consolidating the existing knowledge in this area, we present a pattern that we coined Joint Learning that provides a solution to the decision-making problem of large, distributed self-adaptive IoT systems. With this pattern, elements use a joint learner to make adaptation decisions for individual elements while yielding reliable communication of the overall network. The pattern is applied to two cases to show that the solutions realize the system goals while operating under uncertainties.