CuCo: Graph Representation with Curriculum Contrastive Learning

Graph-level representation learning is to learn low-dimensional representation for the entire graph, which has shown a large impact on real-world applications. Recently, limited by expensive labeled data, contrastive learning based graph-level representation learning attracts considerable attention. However, these methods mainly focus on graph augmentation for positive samples, while the effect of negative samples is less explored. In this paper, we study the impact of negative samples on learning graph-level representations, and a novel curriculum contrastive learning framework for self-supervised graph-level representation, called CuCo, is proposed. Specifically, we introduce four graph augmentation techniques to obtain the positive and negative samples, and utilize graph neural networks to learn their representations. Then a scoring function is proposed to sort negative samples from easy to hard and a pacing function is to automatically select the negative samples in each training procedure. Extensive experiments on fifteen graph classification real-world datasets, as well as the parameter analysis, well demonstrate that our proposed CuCo yields truly encouraging results in terms of performance on classification and convergence.

Learn from Concepts: Towards the Purified Memory for Few-shot Learning

Human beings have a great generalization ability to recognize a novel category by only seeing a few number of samples. This is because humans possess the ability to learn from the concepts that already exist in our minds. However, many existing few-shot approaches fail in addressing such a fundamental problem, {\it i.e.,} how to utilize the knowledge learned in the past to improve the prediction for the new task. In this paper, we present a novel purified memory mechanism that simulates the recognition process of human beings. This new memory updating scheme enables the model to purify the information from semantic labels and progressively learn consistent, stable, and expressive concepts when episodes are trained one by one. On its basis, a Graph Augmentation Module (GAM) is introduced to aggregate these concepts and knowledge learned from new tasks via a graph neural network, making the prediction more accurate. Generally, our approach is model-agnostic and computing efficient with negligible memory cost. Extensive experiments performed on several benchmarks demonstrate the proposed method can consistently outperform a vast number of state-of-the-art few-shot learning methods.

Societal context-dependent multi-modal transportation network augmentation in Johannesburg, South Africa

In most developing countries, formal and informal transportation schemes coexist without effective and smart integration. In this paper, the authors show how to leverage opportunities offered by formal and informal transportation schemes to build an integrated multi-modal network. Precisely, the authors consider integration of rickshaws to a bus-train network, by taking into account accessibility and societal constraints. By modelling the respective networks with weighted graphs, a graph augmentation problem is solved with respect to a composite cost taking into account constraints on the use of rickshaws. The solution, is based on finding a minimum cost spanning tree of a merged graph. The method is applied in the South African context, in the city of Johannesburg where rickshaws are not yet a significant part of the transportation system. The implications of the study reveal that using non-motorised transportation services is a viable option of improving mobility in the city. The composite cost introduced herein could be used for new routing algorithm including societal, environmental, architectural contexts and commuter experiences through rating.

Population-specific genome graphs improve high-throughput sequencing data analysis: A case study on the Pan-African genome

ABSTRACTGraph-based genome reference representations have seen significant development, motivated by the inadequacy of the current human genome reference for capturing the diverse genetic information from different human populations and its inability to maintain the same level of accuracy for non-European ancestries. While there have been many efforts to develop computationally efficient graph-based bioinformatics toolkits, how to curate genomic variants and subsequently construct genome graphs remains an understudied problem that inevitably determines the effectiveness of the end-to-end bioinformatics pipeline. In this study, we discuss major obstacles encountered during graph construction and propose methods for sample selection based on population diversity, graph augmentation with structural variants and resolution of graph reference ambiguity caused by information overload. Moreover, we present the case for iteratively augmenting tailored genome graphs for targeted populations and test the proposed approach on the whole-genome samples of African ancestry. Our results show that, as more representative alternatives to linear or generic graph references, population-specific graphs can achieve significantly lower read mapping errors, increased variant calling sensitivity and provide the improvements of joint variant calling without the need of computationally intensive post-processing steps.

Adding Edges for Maximizing Weighted Reachability

In this paper, we consider the problem of improving the reachability of a graph. We approach the problem from a graph augmentation perspective, in which a limited set size of edges is added to the graph to increase the overall number of reachable nodes. We call this new problem the Maximum Connectivity Improvement (MCI) problem. We first show that, for the purpose of solve solving MCI, we can focus on Directed Acyclic Graphs (DAG) only. We show that approximating the MCI problem on DAG to within any constant factor greater than 1 − 1 / e is NP -hard even if we restrict to graphs with a single source or a single sink, and the problem remains NP -complete if we further restrict to unitary weights. Finally, this paper presents a dynamic programming algorithm for the MCI problem on trees with a single source that produces optimal solutions in polynomial time. Then, we propose two polynomial-time greedy algorithms that guarantee ( 1 − 1 / e ) -approximation ratio on DAGs with a single source, a single sink or two sources.