To leverage rich data distributed at the network edge, a new machine-learning paradigm, called edge learning, has emerged where learning algorithms are deployed at the edge for providing intelligent services to mobile users.While computing speeds are advancing rapidly, the communication latency is becoming the bottleneck of fast edge learning. To address this issue, this work is focused on designing a low-latency multi-access scheme for edge learning. To this end, we consider a popular privacy-preserving framework, federated edge learning (FEEL), where a global AI-model at an edge-server is updated by aggregating (averaging) local models trained at edge devices. It is proposed that the updates simultaneously transmitted by devices over broadband channels should be analog aggregated "over-the-air" by exploiting the waveform-superposition property of a multi-access channel.Such broadband analog aggregation (BAA) results in dramatical communication-latency reduction compared with the conventional orthogonal access (i.e., OFDMA). In this work, the effects of BAA on learning performance are quantified targeting a single-cell random network. First, we derive two tradeoffs between communication-andlearning metrics, which are useful for network planning and optimization. The power control ("truncated channel inversion") required for BAA results in a tradeoff between the update-reliability [as measured by the receive signalto-noise ratio (SNR)] and the expected update-truncation ratio. Consider the scheduling of cell-interior devices to constrain path loss. This gives rises to the other tradeoff between the receive SNR and fraction of data exploited in learning. Next, the latency-reduction ratio of the proposed BAA with respect to the traditional OFDMA scheme is proved to scale almost linearly with the device population. Experiments based on a neural network and a real dataset are conducted for corroborating the theoretical results. In addition, we discuss the extensions of BAA to acquire safety against adversary attacks and integrate beamforming for enhancing cell-edge links.

The recent revival of artificial intelligence (AI) is revolutionizing almost every branch of science and technology. Given the ubiquitous smart mobile gadgets and Internet of Things (IoT) devices, it is expected that a majority of intelligent applications will be deployed at the edge of wireless networks. This trend has generated strong interests in realizing an "intelligent edge" to support AI-enabled applications at various edge devices. Accordingly, a new research area, called edge learning, emerges, which crosses and revolutionizes two disciplines: wireless communication and machine learning. A major theme in edge learning is to overcome the limited computing power, as well as limited data, at each edge device. This is accomplished by leveraging the mobile edge computing (MEC) platform and exploiting the massive data distributed over a large number of edge devices. In such systems, learning from distributed data and communicating between the edge server and devices are two critical and coupled aspects, and their fusion poses many new research challenges. This article advocates a new set of design principles for wireless communication in edge learning, collectively called learning-driven communication. Illustrative examples are provided to demonstrate the effectiveness of these design principles, and unique research opportunities are identified. ! 1Compared with cloud and on-device learning, edge learning has its unique strengths. First, it has the most balanced resource support (see Fig. 1), which helps achieving the best tradeoff between the AI-model complexity and the model-training speed. Second, given its proximity to data sources, edge learning overcomes the drawback of cloud learning that fails to process real-time data due to excessive propagation delay and also network congestion caused by uploading data to the cloud. Furthermore, the proximity gives an additional advantage of location-and-context awareness. Last, compared with on-device learning, edge learning achieves much higher learning accuracy by supporting more complex models and more importantly aggregating distributed data from many devices. Due to the all-rounded capabilities, edge learning can support a wide spectrum of AI models to power a broad range of mission-critical applications, such as autodriving, rescue-operation robots, disaster avoidance and fast industrial control. Nevertheless, edge learning is at its nascent stage and thus remains a largely uncharted area with many open challenges. Fig. 1. Layered in-network machine learning architecture.The main design objective in edge learning is the fast intelligence acquisition from the rich but highly distributed data at subscribed edge devices. This critically depends on data processing at edge servers, as well as efficient communication between edge servers and edge devices. Compared with increasingly high processing speeds at edge servers, communication suffers from hostility of wireless channels (e.g., pathloss, shadowing, and fading), and consequently forms the bottleneck for ultra...

In future Internet-of-Things (IoT) networks, sensors or even access points can be mounted on ground/aerial vehicles for smart-city surveillance or environment monitoring. For such high-mobility sensing, it is impractical to collect data from a large population of sensors using any traditional orthogonal multi-access scheme as it would lead to excessive latency. To tackle the challenge, a technique called over-the-air computation (AirComp) was recently developed to enable a data-fusion to receive a desired function (e.g., averaging or geometric mean) of sensing data from concurrent sensor transmissions. This is made possible by exploiting the superposition property of a multiaccess channel. Targeting a multi-antenna sensor network, this work aims at developing multiple-input-multiple output (MIMO) AirComp for enabling high-mobility multi-modal sensing where a multi-modal sensor monitors multiple environmental parameters such as temperature, pollution and humidity. To be specific, we design MIMO-AirComp equalization and channel feedback techniques for spatially multiplexing multi-function computation, each corresponding to a particular sensing-data type. Given the objective of minimizing sum mean-squared error via spatial diversity, a close-to-optimal equalizer is derived in closed-form using differential geometry. The solution can be computed as the weighted centroid of points (subspaces) on a Grassmann manifold, where each point represents the subspace spanned by the channel coefficient matrix of a sensor. As a by-product, the problem of MIMO-AirComp equalization is proved to have the same form as the classic problem of multicast beamforming, establishing the AirComp-multicasting duality. Its significance lies in making the said Grassmannian-centroid solution method transferable to the latter problem which otherwise is solved using the more computation-intensive semidefinite relaxation method in the literature. Last, building on the AirComp equalization solution, an efficient channel-feedback technique is designed for an access point to receive the equalizer from simultaneous sensor transmissions of designed signals that are functions of local channel-state information. This overcomes the difficulty of provisioning orthogonal feedback channels for many sensors.