Multi-relation graphs intuitively capture the heterogeneous correlations among real-world entities by allowing multiple types of relationships to be represented as entity-connecting edges, i.e., two entities could be correlated with more than one type of relationship. This is important in various applications such as social network analysis, ecology, and bio-informatics. Existing studies on these graphs usually consider an edge label constraint perspective, where each edge contains only one label and each edge is considered independently. For example, there are lines of research focusing on reachability between two vertices under a set of edge label constraints, or finding paths whose consecutive edge labels satisfy a user-specified logical expression. This is too restricted in real graphs, and in this work, we define a generic correlation constraint on multi-relation graphs from the perspective of vertex correlations, where a correlation can be defined recursively. Specifically, we formalize and investigate the shortest path problem over large multi-relation graphs in the presence of both necessity and denial constraints, which have various real applications. We show that it is nontrivial to apply conventional graph traversal algorithms (e.g., BFS or DFS) to address the challenge. To effectively reduce the search space, we propose a Hybrid Relation Encoding method, a.k.a. HyRE, to encode both topological and relation information in a compact way. We conduct extensive experiments over large real-world graphs to validate the effectiveness and efficiency of the proposed solution.

Performing online selective queries against graphs is a challenging problem due to the unbounded nature of graph queries which leads to poor computation locality. It becomes even difficult when a graph is too large to be fit in the memory. Although there have been emerging efforts on managing large graphs in a distributed and parallel setting, e.g., Pregel, HaLoop and etc, these computing frameworks are designed from the perspective of scalability instead of the query efficiency. In this work, we present our solution methodology for online selective graph queries based on the shortest path distance semantic, which finds various applications in practice. The essential intuition is to build a distance-aware index for online distance-based query processing and to eliminate redundant graph traversal as much as possible. We discuss how the solution can be applied to two types of research problems, distance join and vertex set bonding, which are distance-based graph pattern discovery and finding the structure-wise bonding of vertices, respectively.

A prevalent limitation of optimizing over a single objective is that it can be misguided, becoming trapped in local optimum. This can be rectified by Quality-Diversity (QD) algorithms, where a population of high-quality and diverse solutions to a problem is preferred. Most conventional QD approaches, for example, MAP-Elites, explicitly manage a behavioral archive where solutions are broken down into predefined niches. In this work, we show that a diverse population of solutions can be found without the limitation of needing an archive or defining the range of behaviors in advance. Instead, we break down solutions into independently evolving species and use unsupervised skill discovery to learn diverse, high-performing solutions. We show that this can be done through gradient-based mutations that take on an information theoretic perspective of jointly maximizing mutual information and performance. We propose Diverse Quality Species (DQS) as an alternative to archive-based QD algorithms. We evaluate it over several simulated robotic environments and show that it can learn a diverse set of solutions from varying species. Furthermore, our results show that DQS is more sample-efficient and performant when compared to other QD algorithms. Relevant code and hyper-parameters are available at: https://github.com/rwickman/NEAT_RL

Durability indicates a system's ability to preserve committed data in the long term. It is one of the key requirements for storage systems. Various fault tolerance strategies have been proposed for traditional distributed databases, peer-to-peer (P2P) systems, and grid systems. These strategies were all developed for specific platforms and application types and have been tailored to the characteristics of the underlying system architectures and application requirements. Cloud systems differ from these previous frameworks in that they are designed to support large numbers of customer-oriented applications, each with different quality of service (QoS) requirements and resource consumption characteristics. Moreover, most cloud architectures are deployed over large-scale geographically distributed infrastructures, raising challenges on data durability as well as system efficiency and scalability. In this paper, we investigated different approaches to the design of durable data cloud platforms. In particular, we consider prevailing cloud platforms and cloud-based applications and examine the impacts of different redundancy and repair strategies to enhance the durability of the data in the cloud. We verify the performance of various approaches to data durability in the Cloud through extensive simulations.