The exploratory visual analysis (EVA) of time series data uses visualization as the main output medium and input interface for exploring new data. However, for users who lack visual analysis expertise, interpreting and manipulating EVA can be challenging. Thus, providing guidance on EVA is necessary and two relevant questions need to be answered. First, how to recommend interesting insights to provide a first glance at data and help develop an exploration goal. Second, how to provide step-by-step EVA suggestions to help identify which parts of the data to explore. In this work, we present a reinforcement learning (RL)-based system, Visail, which generates EVA sequences to guide the exploration of time series data. As a user uploads a time series dataset, Visail can generate step-by-step EVA suggestions, while each step is visualized as an annotated chart combined with textual descriptions. The RL-based algorithm uses exploratory data analysis knowledge to construct the state and action spaces for the agent to imitate human analysis behaviors in data exploration tasks. In this way, the agent learns the strategy of generating coherent EVA sequences through a well-designed network. To evaluate the effectiveness of our system, we conducted an ablation study, a user study, and two case studies. The results of our evaluation suggested that Visail can provide effective guidance on supporting EVA on time series data.

Space-air-ground integrated network (SAGIN) is a prominent architecture for the future wireless communication system. Due to the long distance transmission of the satellite communication and the limited battery capacity of aircrafts, the energy cost has become one of the dominant problems of SAGIN. Mobile edge computing (MEC) offers a potential solution by offloading partial tasks to nodes with higher computational capability. In this paper, we study a hybrid offloading problem where both unmanned aerial vehicles (UAVs) and ground users have tasks to be processed, and UAVs and the satellite can provide offloading service. The aim is to minimize the system energy consumption under time delay constraint by choosing the optimal offloading objects and jointly optimizing the offloading proportion and computing resource allocation. The optimization problem is highly nonconvex and difficult to solve optimally. To tackle the problem, we first derive the closed-form solution of computation resources allocation and further propose a low-complex algorithm based on successive convex approximation (SCA). Simulation results show that the proposed hybrid of-floading scheme can significantly reduce energy consumption compared to benchmark schemes. Moreover, by increasing the transmission power of users and UAVs in a certain range, total energy consumption can be effectively reduced. And increasing the number of UAVs can improve the energy efficiency.