A massive amount of data is generated at edge networks with emerging self-driving cars, drones, robots, smartphones, wireless sensors, smart meters, and health monitoring devices. This vast data is expected to be processed via artificial intelligence and machine learning (AI/ML) algorithms, which is extremely challenging over resource-constrained edge devices. The first part of the talk will focus on model-distributed inference (MDI), which advocates that an ML model is partitioned and distributed across multiple devices. These partitions are processed in parallel to reduce the ML inference time. In this context, we will present our adaptive, resilient, multi-source, and privacy-preserving MDI solutions. The second part of the talk will present our work on communication-efficient distributed and decentralized federated learning at the edge.