MLSNews

《未闻》区块链技术日报,聚焦业内技术更新,做您最好的眼睛。

View project on GitHub

​MLSNEWS

——————◆ 20190624 ◆——————

  • 第2层方案概述:Plasma,状态通道,侧链,Roll Ups

    随着区块链的日益普及,现代协议的容量仍然非常有限,出现了许多所谓的第2层协议,如状态通道,侧链,等离子和Roll Ups。

    任何第2层解决方案背后的核心思想是,它允许多方以某种方式安全地进行交互,而无需在主链(即第1层)上发布交易,但仍在某种程度上利用主链的安全性作为仲裁者。

    • 状态通道

    作为第2层解决方案的一个很好的介绍性示例,我们首先考虑简单的支付渠道。支付渠道是当今最常采用的第2层解决方案之一。例如,Lightning Network基于支付渠道。

    • 侧链

    一个简单的侧链背后的核心思想是拥有一个完全独立的区块链,它有自己的验证器和操作符,它具有从主链传输资产的桥梁,并可选择将块头快照到主链以防止分叉。

    • Plasma

    Plasma是一种能够实现“非监禁”侧链的结构,也就是说,即使所有侧链(通常称为“Plasma链”)验证器都串通进行任何类型的对抗行为,Plasma链上的资产也是安全的,并且可以是退出了主链。

    • Roll Ups

    Roll Up实际上是一个侧链,从某种意义上说它产生块,并将这些块快照到主链。但是,Roll Up中的运营商不受信任。因此,假设操作员可以在任何时候尝试停止生成块,产生无效块,隐藏数据或尝试其他一些对抗行为。

    With the increasing adoption of blockchains, and still very limited capacity of modern protocols, many so-called Layer 2 protocols emerged, such as State Channels, Side Chains, Plasma and Roll Ups. This blog post dives relatively deeply into the technical details of each approach and their benefits and disadvantages.

    The core idea behind any Layer 2 solution is that it allows several parties to securely interact in some way without issuing transaction on the main chain (which is the Layer 1), but still to some extent leveraging the security of the main chain as the arbitrator.

    • State Channels

    As a good introductory example of Layer 2 solutions, let’s first consider simple payment channels. Payment channels are one of the most adopted Layer 2 solutions today. Lightning Network, for example, is based on Payment Channels.

    • Side Chains

    The core idea behind a simple side chain is to have a completely separate blockchain, with its own validators and operators, that has bridges to transfer assets to and from the main chain, and optionally snapshots the block headers to the main chain to prevent forks.

    • Plasma

    Plasma is a construction that enables “non-custodial” sidechains, that is, even if all sidechain (commonly called “plasma chain”) validators collude to conduct any type of adversarial behavior, the assets on the plasma chain are safe, and can be exited to the mainchain.

    • Roll Ups

    blocks, and snapshots those blocks to the main chain. The operators in the Roll Up, however, are not trusted. Thus it is assumed that at any point the operators can attempt to stop producing blocks, produce an invalid block, withhold data, or attempt some other adversarial behavior.

  • 以太坊隐私交易方案 Hopper 开源代码,发布移动应用内测版

    以太坊隐私交易方案 Hopper 宣布完全开源代码,并发布 iOS 端移动应用内测版。Hopper 使用零知识证明(ZKP)技术「混合」用户交易,用户可将 ETH 存入 Hopper 智能合约,而从另一个账户提取资金。目前用户一次仅可存入 1 枚 ETH 以保持匿名,Hopper 还提醒由于应用尚处于内测阶段,没有进行完全审计,因而资金有丢失的风险

    Etaifang Privacy Trading Scheme Hopper announced full open source code, and released the iOS mobile application internal beta version. Hopper uses zero knowledge proof (ZKP) technology to “mix” user transactions. Users can deposit ETH in Hopper smart contracts and withdraw funds from another account. At present, users can only access one ETH at a time to keep anonymity. Hopper also reminds that because the application is still in the internal testing stage and has not been fully audited, there is a risk of loss of funds.

  • 量化MXnet中的神经网络模型以实现区块链上的严格一致性

    在各种平台和设备上部署深度学习模型非常有意义。深度神经网络不仅用于超级计算机或GPU云,而且在资源受限的设备或边缘设备中也越来越多地被采用。与超级计算机不同,边缘设备具有低计算能力,具有存储器和能量消耗限制。在区块链的新兴领域,它要求计算的每个步骤都是严格确定的并且资源消耗最少,因此神经网络部署起来更加困难。

    随着分布式总账技术的发展,出现了对更复杂计算的需求。浮点算法和并行计算的非确定性本质(例如,对一系列浮点数求和)对于云计算环境中的DNN模型应该不是问题,因为DNN模型推断的目的通常是为了更好的精度,不适用于未来的审计或比特级比较。然而,在区块链的情况下,不确定性结果是不合需要的,其中不同节点需要在最终确定区块链之前验证交易以达成共识。每个节点都有自己的硬件规范,运行不同版本的操作系统,这使得计算标准化变得更加困难。换句话说,每次运行模型,输入相同,在异构计算系统中必须产生位级相同的结果。

    There is significant interest in deploying deep learning models on various platforms and devices. Deep neural networks are not just being used in supercomputers or GPU clouds, but are also experiencing increased adoption in resource-constrained devices, or edge devices. Unlike supercomputers, edge devices have low computing power with memory and energy consumption constraints. In the emerging field of blockchain, where it demands every step of computation to be strictly deterministic and have minimal resource consumption, it is even harder for a neural network to deploy.

    As the technology of distributed ledger develops, needs for more complex computations emerge. The nondeterministic nature of floating-point arithmetic and parallel computation, e.g., summation over a series of floating-point numbers, should not be a problem for DNN models in a cloud computing environment, since the purpose of DNN model inference is often for better precision, not for future auditing or bit-level comparison. However, the nondeterministic result is undesirable in the occasion of blockchain, where different nodes need to verify the transactions to reach consensus before finalizing on the blockchain. Each node has its own hardware specification running different versions of operating systems, making it harder to standardize the computation. In other words, every single run of a model, with the same input, in heterogeneous computing systems must yield bit-level identical results.