The intrinsic non-stationarity of urban spatiotemporal (ST) streams, particularly unique distribution shifts that evolve over time, poses substantial challenges for accurate urban ST forecasting. Existing works often overlook these dynamic shifts, limiting their ability to adapt to evolving trends effectively. To address this challenge, we propose DOL, a novel Distribution-aware Online Learning framework designed to handle the unique shifts in urban ST streams. DOL introduces a streaming update mechanism that leverages streaming memories to strategically adapt to gradual distribution shifts. By aligning network updates with these shifts, DOL avoids unnecessary updates, reducing computational overhead while improving prediction accuracy. DOL also incorporates an adaptive spatiotemporal network with a location-specific learner, enabling it to handle diverse urban distribution shifts across locations. Experimental results on four real-world datasets confirm DOL's superiority over state-of-the-art models. The source code is available at https://github.com/cwang-nus/DOL.

Forecasting time series values based on historic covariates has been an active area of research in statistics and machine learning. With the availability of computation resources and big data infrastructure supporting massive volume, velocity and variety, the algorithms have evolved from classic statistical learning to neural-network driven loss minimisation techniques. While state of the art attention and self-attention-transformers have shown promise of improved performance with sufficient training data, most of them fail to generalise to different problems of time-series modelling (such as classification and extremum forecasting) with asynchronously sampled covariates. This paper introduces the concept of a generalised time series embedding and transfer learning for time series (analogous to token-to-vector or image-to-vector embeddings in language and vision models respectively) that allow joint training with a unified interface. The major benefit of this work is a unified embedding model employing multi-attention for feature representation which enables benchmark performance against state of the art models from recent literature.

To serve advanced use-cases in industrial internet of things (IIoT) setups, communication and computation over wireless networks have faced overlapping resource management challenges. Two crucial resources in this context are radio re-sources and computational resources. The problem to achieve the ultra-low latency for mission critical applications is motivating enterprises to invest in offloading capability of computation heavy tasks while retaining the bandwidth efficiency of edge nodes. This work proposes a novel multi-hop offloading framework powered by deep reinforcement learning to aid the edge nodes in making intelligent decisions on task offloading. The proposed method is benchmarked against existing state of the art techniques to measure task completion delay and algorithmic runtime.

Rise in wireless data traffics requires innovative interference management techniques to meet the demand. Transmit cooperation among multiple base stations has been proposed as a solution to this challenge and improve throughput. For base stations equipped with massive MIMO arrays, a proposed subspace encoding technique can decouple the joint multicell resource allocation problems into independent problems for each cell. The subspace encoding in the multicell environment is analytically equivalent to a reduction of degrees of freedom in a single cell broadcast channel. Hence, the resulting computational and conceptual simplicity can be exploited to arrive at the optimal resource allocation for maximum throughput.

Traditional cellular systems have depended on coverage areas partitioned into disjoint cells where each user is served by a single base station, without any cooperation. To deal with the resulting inter-cell interference, this work combines the advantages of CoMP along with detection and broadcasting techniques with different channel output suitable for massive MIMO. Two novel interference management techniques, viz. Centralised Interference Cancellation (CIC) and Distributed Interference Cancellation (DIC) to optimise power allocation in a cooperative framework have been proposed. The schemes point out the trade off between achievable performance and computational complexity. Since large number of antennas require mathematical manipulation of big random matrices, a suboptimal two layer decoding method is proposed which combines selection diversity with least mean square filtering.

Duality relations among achievable QoS parameters help us transform the problems in the broadcast channel to convex problems in multiple access channels. But a multiple access channel being considered as dual of a broadcast channel must exploit the additional freedom of power allocation traditionally considered only in a broadcast channel. This work proposes a novel algorithm for optimising the sum-capacity of a multiple access channel under sum power constraint where the users are coupled by individual power usage as well as interference. Optimal resource allocation has a certain distributed feature which can be exploited to formulate the algorithm. Power allocation problem for an arbitrary number of users is solved by dealing with only two users at a time. The algorithm gives insights about how the decoding priority affects achievable rates of different users which have been used for further characterisation of a rate region boundary. Simulation results have been presented in the end to support the theoretical arguments.