Sharding is a promising solution to enhance the scalability of blockchain. However, previous sharding systems adopt the lock-based cross-shard protocol to exclusively handle one-shot cross-shard transactions, leading to low-efficiency executions and unavailable calls when handling complex cross-shard contracts that introduce multi-shot cross-shard transactions to invoke multiple contracts managed by different shards.In this paper, we aim to enable efficient execution of arbitrarily complex cross-shard contracts in blockchain sharding systems. First, we perform a calling-flow analysis on Ethereum contracts with more than 180 million real-world transactions and find that about 30% transactions invoke complex contracts. Then, motivated by the properties of these complex contracts, we propose an off-chain execution model, called ShardCon, to achieve efficient executions for complex cross-shard contracts by decoupling the contract execution from the cross-shard consensus. Next, we introduce a cross-shard contract execution engine and a contract-driven deployment rule to the overheads introduced by off-chain executions. Moreover, to adapt to the multi-chain property of a sharding system, we introduce an off-chain state atomic commit protocol. Finally, we implement a prototype and evaluate it with concrete cross-shard contracts, showing that ShardCon can achieve more than 10x increase in throughput and 2x decrease in confirmation latency than the state-of-the-art sharding systems.

Building on top of blockchain, payment channel networks-backed (PCNs) cryptocurrencies emerge as a promising solution for a mobile payment system with fewer intermediaries, more security, higher speed, and lower cost. A key problem for PCN is payment channel rebalancing, that is, finding a set of circular transactions that restore a PCN with skewed channel balances back into an equilibrium state. However, existing practice on payment channel rebalancing either has a hard limit on the problem size or tends to fall into local optimum. To address these challenges, we propose DRL-PCR, aDeepReinforcementLearning-basedPaymentChannelRebalancing algorithm. On one hand, DRL-PCR leverages the strong approximation ability of deep neural networks to handle large problem spaces. On the other hand, DRL-PCR decomposes the rebalancing problem into a sequence of decision-making problems and progressively builds the final solution. By aiming to find a globally optimized solution and solving the long-term optimization model of DRL, DRL-PCR is superior to greedy-based algorithms and can mitigate the risk of getting trapped in a local optimum. In particular, payment channel rebalancing typically involves dealing with graph-structured data, where the major obstacle lies in understanding the sophisticated circular dependencies between payment channels and routing paths. DRL-PCR achieves this by encoding the input data with a novel graph neural network-based model and capturing the circular dependencies through a customized message passing process. In addition, considering the distributed nature of PCN, DRL-PCR uses a leadership election protocol to elect leaders for decision-making. Evaluations on the historical data of two real-world PCNs demonstrate that DRL-PCR can restore the PCN to a more balanced state and improve the transaction throughput and success ratios by up to 2.1x and 1.6x, respectively.

The growing maturity of single-cell RNA-sequencing (scRNA-seq) technology allows us to explore the heterogeneity of tissues, organisms, and complex diseases at cellular level. In single-cell data analysis, clustering calculation is very important. However, the high dimensionality of scRNA-seq data, the ever-increasing number of cells, and the unavoidable technical noise bring great challenges to clustering calculations. Motivated by the good performance of contrastive learning in multiple domains, we propose ScCCL, a novel self-supervised contrastive learning method for clustering of scRNA-seq data. ScCCL first randomly masks the gene expression of each cell twice and adds a small amount of Gaussian noise, and then uses the momentum encoder structure to extract features from the enhanced data. Contrastive learning is then applied in the instance-level contrastive learning module and the cluster-level contrastive learning module, respectively. After training, a representation model that can efficiently extract high-order embeddings of single cells is obtained. We selected two evaluation metrics, ARI and NMI, to conduct experiments on multiple public datasets. The results show that ScCCL improves the clustering effect compared with the benchmark algorithms. Notably, since ScCCL does not depend on a specific type of data, it can also be helpful in clustering analysis of single-cell multi-omics data.