Federated systems enable collaborative training on highly heterogeneous, non-i.i.d. data through model personalization, which can be facilitated by employing multi-task learning. However, multi-task learning algorithms are often implemented using methods like stochastic gradient descent, which may suffer from slow convergence in a multi-task federated setting. To accelerate the training procedure, we design an efficient iterative distributed method based on the alternating direction method of multipliers (ADMM) for support vector machines (SVMs), which tackles federated classification and regression. The proposed method utilizes efficient computations and model exchange in a network of heterogeneous nodes and allows personalization of the learning model in the presence of non-i.i.d. data. To ensure data privacy, we introduce a randomization algorithm that helps avoid data inversion. Finally, we analyze the impact of the proposed privacy mechanisms and participant hardware and data heterogeneity on the system performance. Our experiments confirm the advantages of the proposed ADMM-based personalized federated multi-task learning.

Federated Learning (FL) enables collaborative training of machine learning models for edge devices (e.g., mobile phones) over a network without revealing raw data of the participants. The existing FL benchmarks mostly assume static distribution of local data over time failing to capture the behavior of real-world applications with space-time varying data (e.g., autonomous cars). Our framework addresses this limitation by leveraging popular open source physics simulator (CARLA) and FL framework (OpenFL) to allow collection of streaming data from the mobile agents and feeding them to a practically deployable FL engine for online collaborative training. It also provides the FL researchers with the ability to model data heterogeneity, annotate data with practically zero cost, and perform reproducible continual FL experiments. We believe that this is one of the first attempts to demonstrate online FL on realistic streaming datasets from a virtual world. The demo showcases this framework using a popular object detection use case.

Most existing deep neural networks are static, which means they can only perform inference at a fixed complexity. But the resource budget can vary substantially across different devices. Even on a single device, the affordable budget can change with different scenarios, and repeatedly training networks for each required budget would be incredibly expensive. Therefore, in this work, we propose a general method called MutualNet to train a single network that can run at a diverse set of resource constraints. Our method trains a cohort of model configurations with various network widths and input resolutions. This mutual learning scheme not only allows the model to run at different width-resolution configurations but also transfers the unique knowledge among these configurations, helping the model to learn stronger representations overall. MutualNet is a general training methodology that can be applied to various network structures (e.g., 2D networks: MobileNets, ResNet, 3D networks: SlowFast, X3D) and various tasks (e.g., image classification, object detection, segmentation, and action recognition), and is demonstrated to achieve consistent improvements on a variety of datasets. Since we only train the model once, it also greatly reduces the training cost compared to independently training several models. Surprisingly, MutualNet can also be used to significantly boost the performance of a single network, if dynamic resource constraints are not a concern. In summary, MutualNet is a unified method for both static and adaptive, 2D and 3D networks. Code and pre-trained models are available at https://github.com/taoyang1122/MutualNet.

Efficient radio resource allocation is a fundamental optimization problem for wireless networks, and has been widely studied in the past. However, wireless systems are evolving to have a much larger parameter space along with richer set of applications and user requirements leading to significant increase in complexity. This paper draws from recent breakthroughs in applying deep reinforcement learning (RL) for control problems containing large state space dimensionality. In particular, a deep RL approach is explored for the problem of allocating time and frequency resources in OFDMA wireless systems to optimize different objective functions using per-station channel quality and traffic information as inputs. Such approaches hold the potential for agents to learn resource allocation and scheduling policies directly from experience rather than using carefully crafted heuristic algorithms based upon models of the environment and stations. Learning directly from experience also means that the policies that result from the online learning should be more robust to imperfect inputs such as noisy, delayed or missing information compared to model-based heuristic approaches. The results in this paper show promise for a deep RL agent using a policy gradient algorithm to learn policies that approach or exceed the performance of well-known model-based approaches such as max-weight and proportional fair policies. The online adaptation algorithms used for the deep RL agent also demonstrate reasonable adaptability and robustness to varying traffic conditions.

Cellular systems are rapidly evolving from a homogeneous macrocell deployment to a heterogeneous deployment of macrocells overlaid with many small cells. The striking features of small cells include small coverage area, ad-hoc deployment, and flexible backhaul connectivity. These features call for a profound rethinking of traditional cellular concepts including mobility management and interference management among others. Owing to the unique features, coordinated small cells or commonly referred to as network of small cells promises several benefits including efficient mobility management in a rapid and scalable fashion. The problem of handover in a high-density small cell deployment is studied in this work. A novel local anchor-based architecture for a static cluster of small cells is proposed using which new handover schemes are presented. Such clusters are prevalent in the evolving cellular systems involving high-density small cell deployments in urban, residential, and enterprise environments. A mathematical framework is developed using discrete-time-Markov-models to evaluate the proposed schemes. Using this, closed-form expressions for key handover metrics including handover cost and interruption time are derived. Extensive numerical and simulation studies indicate significant savings of over 50 percent in the handover costs and, more importantly, up to 80 percent in the handover interruption time compared to the existing 3GPP scheme for coordinated small cells.

Smallcell technology is gaining significance as part of the next-generation cellular systems due to their performance benefits in terms of increased network capacity and improved indoor and local coverage. Hybrid access smallcells, which provide service to both indoor as well as neighboring users, adopt adhoc policies to guarantee performance benefits to indoor home users in the presence of external neighboring users. Such policies must be able to stabilize user queues as well as to provision performance benefits in terms of delay and throughput, especially for the indoor users. As a result, classification of user data in terms of traffic type and user type is required to effectively achieve the differentiated QoS performance. In this paper, a traffic-aware utility function is proposed, which takes into account for the user's priority index and traffic characteristics to efficiently provide differentiated QoS benefits to users served under an OFDMA hybrid smallcell. The problem of the traffic-aware utility based scheduling under power constraints is posed as an optimization objective and an optimal algorithm for the scheduling problem is presented. The results show that the proposed scheme achieves QoS performance benefits in terms of throughput and delay.

Relay is one of the key features being considered for IMT-Advanced systems. The relay architectures defined in IEEE 802.16m and 3GPP LTE-Advanced are optimized only for non-mobile relay, i.e., the Relay Station is attached to a designated Base Station and becomes a part of the fixed access network. A mobile relay architecture, where relay may switch attached BS according to operation demand, will promise more resilient relay deployment. In this paper, we first highlight three use cases where mobile relays can offer useful deployment options. Next, we propose an enhanced handover mechanism for relay-based group mobility by extending the IEEE 802.16m specification on relay. To this end, we describe the handover architecture for mobile relay highlighting the C-plane and U-plane enhancements required in order to support group mobility of mobile stations. Using illustration and comparisons, we show that mobile relay will offer significant benefits of signaling overhead reduction and provide users seamless mobility experience compared to fixed relays.