Deep Neural Networks (DNNs) have been widely deployed in mobile applications. DNN inference latency is a critical metric to measure the service quality of those applications. Collaborative inference is a promising approach for latency optimization, where partial inference workloads are offloaded from mobile devices to cloud servers. Model partition problems for collaborative inference have been well studied. However, little attention has been paid to optimizing offloading pipeline for multiple DNN inference jobs. In practice, mobile devices usually need to process multiple DNN inference jobs simultaneously. We propose to jointly optimize the DNN partitioning and pipeline scheduling for multiple inference jobs. We theoretically analyze the optimal scheduling conditions for homogeneous chain-structure DNNs. Based on the analysis, we proposed near-optimal partitioning and scheduling methods for chain-structure DNNs. We also extend those methods for general-structure DNNs. In addition, we extend our problem scenario to handle heterogeneous DNN inference jobs. A layer-level scheduling algorithm is proposed. Theoretical analyses show that our proposed method is optimal when computation graphs are tree-structure. Our joint optimization methods are evaluated in a real-world testbed. Experiment results show that our methods can significantly reduce the overall inference latency of multiple inference jobs compared to partition-only or schedule-only approaches.

Due to the additive property of most machine learning objective functions, the training can be distributed to multiple machines. Distributed machine learning is an efficient way to deal with the rapid growth of data volume at the cost of extra inter-machine communication. One common implementation is the parameter server system which contains two types of nodes: worker nodes, which are used for calculating updates, and server nodes, which are used for maintaining parameters. We observe that inefficient communication between workers and servers may slow down the system. Therefore, we propose a graph partition problem to partition data among workers and parameters among servers such that the total training time is minimized. Our problem is NP-Complete. We investigate a two-step heuristic approach that first partitions data, and then partitions parameters. We consider the trade-off between partition time and the saving in training time. Besides, we adopt a multilevel graph partition approach to fit the bipartite graph partitioning. We implement both approaches based on an open-source parameter server platform-PS-lite. Experiment results on synthetic and real-world datasets show that both approaches could significantly improve the communication efficiency up to 14 times compared with the random partition.

Bike-sharing systems usually suffer from out-of-service events due to bike underflow or overflow. We propose to recruit workers to rebalance station loads. We partition the complex rebalancing problem in temporal and spatial domains. The temporal domain is divided into a sequence of slices with a fixed duration. In each slice, we allocate a pair of overflow/underflow stations to a worker such that the cost is minimized, which is NP-hard. A 3-approximation algorithm is proposed. We further investigate the worker shortage case and extend the matching algorithm to consider the number of unsatisfied users. Then, the configuration dynamic in the sequence of slices is captured by determining the rebalancing target for each rebalancing operation. We investigate heuristic approaches to minimize the total number of bike movements. Furthermore, we extend our scheme to dockless BSSs using clustering techniques. We simulate our algorithms on both real-world and synthetic datasets. Experiment results show that our approaches can reduce the average total detour per slice. In worker shortage, considering the number of unsatisfied users could improve the long-term performance of rebalancing. Besides, we find that our scheme could maintain worker satisfaction over multiple time slices, which indicates the sustainability of our rebalancing scheme.

Nowadays, large cities face numerous problems caused by the rapidly increasing number of vehicles on road. Carpooling has been shown as an efficient solution to ease the traffic pressure. The objective of our carpool scheduling problem is to minimise the number of carpools needed to transport users. Previous research on the similar topic introduces static capacity constraint which limits the carpool size to the vehicle's capacity. However, it is unnecessary since a seat in the vehicle can be occupied by several passengers if their routes are not overlapped. In this paper, we remove this constraint, which allows a vehicle to serve more users than its capacity. We propose a greedy approach based on the iterative matching and merging. Specifically, starting from a set of single-user carpools, the algorithm iteratively checks the merge-ability between carpools, and then applies the maximum matching algorithm to maximise the number of carpools merged. In addition, two methods are proposed to reduce the time complexity of merge-ability checking process. Furthermore, we improve the time efficiency of our algorithms by exploiting geometry properties. Experimental results on synthetic and real-world datasets show that our algorithms have better performances than baseline. GRAPHICAL ABSTRACT