Free and easy week review 丨 AI algorithm cross-border participation in Bitcoin mining? Ethereum successfully completes Istanbul upgrade

Introduction: In this week's academic paper, we will share two studies related to the topic of mining strategies, and they both involve AI algorithms. In the weekly selection of hard core technical articles, we will also see the violence test of 12 hardware wallets, the Bitcoin new technology stack, Zcash zero-knowledge proof system, Ethereum difficulty bomb, and the latest research and development summary of Ethereum.

In addition, Ethereum successfully completed the Istanbul upgrade this weekend. What changes will it undergo in the next period? 45

(Image source:

The following is a selection review of last week's content, enjoy ~

I. Topic Selection of Academic Papers in a Week: How AI Algorithms Are Applied to Blockchain Mining

We know that incentive mechanisms are at the core of unlicensed blockchains (public chains): they motivate participants to run and protect basic consensus protocols.

However, designing incentives that are compatible with incentives is actually very challenging. Specifically, users are either Byzantine or honest, and systems with strong theoretical security guarantees often exclude the analysis of rational users, and they may deviate from honest behavior due to incentives.

Therefore, most of the incentive mechanisms used by public chains today are not absolute, and many of them have not been tested.

Miners invest computing resources to solve the PoW problem. In the early days, people believed that the most profitable mining strategy was honest mining. Once the miner solved the PoW problem, he would broadcast the newly generated block as soon as possible.

Later, in 13 years, Professor Emin Gün Sirer and Assistant Professor Ittay Eyal of Cornell University proposed a mining strategy called selfish mining. Under certain conditions, this strategy can achieve a ratio of Higher returns from honest mining. Since then, research on incentive attacks has sprung up.

Miners who use selfish mining will not immediately broadcast the blocks they mine. They secretly connect their future mined blocks with their detained blocks, and then perform a block withholding attack.

By 2015, researchers at Microsoft Research and Hebrew University of Jerusalem, Israel, Ayelet Sapirshtein and others described the mining problem as a general Markov decision process (MDP) with a large state behavior space. However, the goal of mining MDP It is not a linear function of reward like standard MDP, so it is impossible to solve mining MDP using standard MDP decoding algorithm. To solve this problem, the researchers first converted a mining MDP with a non-linear target into an MDP with a linear target, and then used a standard MDP decoding algorithm on this MDP to find the optimal mining strategy.

And this method, before setting up the MDP, it is necessary to know the various parameter values. In a real blockchain network, accurate parameter values ​​are difficult to obtain and may change over time, thereby hindering them. To the practical adoption of this solution.

Paper 1: " When Blockchain Meets AI: Optimal Mining Strategies Based on Machine Learning "

Professor Liu Shaoqiang of the Chinese University of Hong Kong and IEEE Fellow (Academician), Wang Taotao, an assistant professor and postdoctoral fellow of Shenzhen University, and Zhang Shengli, a professor and doctoral supervisor of Shenzhen University, recently presented a new research paper.

Paper link:

In this paper entitled “When the Blockchain Meets AI: Optimal Mining Strategies Based on Machine Learning”, the researchers used a reinforcement learning (RL) algorithm to dynamically learn a performance by observing and interacting with the network. A mining method that is close to the optimal mining strategy.

Reinforcement learning (RL) algorithms are a machine learning paradigm in which agents learn successful strategies and obtain the greatest long-term returns from trial and error with the environment.

At present, Q-learning is the most popular reinforcement learning (RL) algorithm. It can learn a good strategy by updating a state action value function without the need for an environmental operation model. The reinforcement learning (RL) algorithm has been successfully applied. In many challenging tasks, such as playing video games, Go, and controlling the movement of robots.

However, the original reinforcement learning (RL) algorithm cannot deal with the non-linear objective function of the mining problem.

Therefore, the authors of this paper propose a new multi-dimensional RL algorithm based on Q-learning, and this algorithm can successfully find the optimal mining strategy.


Simulation experiments show that mining through this reinforcement learning (RL) algorithm is more profitable than traditional selfish mining and honest mining.

Free and easy comments: This is the latest example of interdisciplinary research. It also reminds the aborigines of the cryptocurrency world that new technology developments will also cause new problems. Should we be too worried about this? No hurry, let's look at the next research paper first.

Paper 2: "SquirRL: Automatic Discovery of Blockchain Incentive Attacks Using Deep Reinforcement Learning"

Researchers from Carnegie Mellon University, Peking University, Cornell Institute of Technology, and Stanford University recently published an article entitled “ SquirRL: Automatic Discovery of Blockchain Incentive Attacks Using Deep Reinforcement Learning Technology '' paper.

Paper link:

The title seems a bit sloppy, which roughly means: use the Deep Reinforcement Learning (Deep-RL) algorithm to realize the automatic discovery of blockchain incentive attacks.

In the previous paper, we mentioned the reinforcement learning (RL) algorithm. What is the deep reinforcement learning (RL) algorithm?

Deep reinforcement learning is a class of reinforcement learning (RL) algorithms using neural network learning strategies.

The Deep Reinforcement Learning (Deep-RL) algorithm is particularly successful in solving problems with the following two properties:

  1. Well-defined rules;
  2. State space is huge (difficult to control);

The blockchain incentive mechanism is in line with the above nature. In fact, the issue of the blockchain incentive mechanism has an additional advantage, that is, the blockchain reward is processed continuously.

Here is a summary of key points from this paper:

  1. This paper proposes a framework for discovering blockchain incentive attacks using Deep Reinforcement Learning (Deep-RL) algorithms. It aims to provide a universal method for blockchain developers to test the vulnerability incentive mechanism, but does not provide theoretical guarantees ( Note: The so-called theoretical guarantee means that the researcher did not use the framework to find relevant attacks in reality, because incentive attacks are difficult to achieve by themselves).
  2. The framework can be applied to various blockchain protocols, and can recover theoretical selfish mining attacks known in the Bitcoin protocol, while also extending the latest results to previously difficult areas (such as multi-agent settings, larger State space and other protocols).
  3. Research shows that in the Bitcoin protocol, as the number of agents increases, the profit of selfish mining and its variant attacks will gradually decrease, which is consistent with the fact that the selfish mining phenomenon has not been found in reality , although It is unclear whether this observation or other external factors are the cause.
  4. Finally, the paper proves that the framework can also be applied to incentive attacks other than selfish mining;

Authors: Charlie Hou, Mingxun Zhou, Yan Ji, Phil Daian, Florian Tramèr, Giulia Fanti, Ari Juels

Some details of the experiment

The researchers used their proposed SquirRL framework to experimentally compare the blockchain incentive mechanisms of Bitcoin, Ethereum, and GHOST. They conducted 100 experiments on each blockchain protocol, each of which contained 10,000 times. State transition and at least 5000 blocks in the main chain, and then divided into a single agent and multiple agents for evaluation.

Here are the four components involved in this experiment:

  1. Honest: Miners who abide by the agreement;
  2. Optimal Selfish Mining (OSM);
  3. SM1: Selfish mining strategy proposed by Professor Emin G¨un Sirer;
  4. RL: the SquirRL system proposed by researchers;

Here is a graph of the experimental results for a single agent:


We can see that when the hash power (or equity) α held by the attacker is less than 25%, SquirRL does not adopt a "selfish mining" strategy, but recovers the honest mining strategy, and the attacker's hash power (or Equity) When α is greater than 25%, SquirRL outperforms other solutions.

Finally, the results obtained by SquirRL show that the classic selfish mining attack against Bitcoin will lose its effectiveness in the presence of multiple attackers.

These results reveal why selfish mining exists in theory, but it can be a bad attack strategy in the real world.

Free and easy comments: It is also the study of blockchain topics using AI algorithms. Unlike the previous paper, the purpose of this research is to identify relevant blockchain incentive attacks, and it also proves that when the network When there are multiple agents (ie, attackers), the related incentive attack effect will be weakened, which means that the network will eventually reach a balanced state, so we do not need to worry too much.

Second, hard core technical articles of the week

2.1 Metal Bitcoin Seed Storage Stress Test (third round)

This week's first hardcore article is from a crypto-punk Jameson Lopp hardware wallet brute force review.

Article link:

In this experiment, the author brutally destroyed 12 hardware wallets:

  1. Heating test : After heating the hardware wallet for 10 minutes at about 2000 degrees Fahrenheit (about 1093.33 degrees Celsius), quickly throw it into a bucket filled with water to simulate a firefighter extinguishing a fire.
  2. Corrosion test : The hardware wallet device is immersed in hydrochloric acid for 12 hours.
  3. Heavy pressure test : A 20-ton hydraulic machine was used to deform the equipment to simulate the collapse of a large building and press it into a hardware wallet;

Easy and Easy Comment: The hardware wallet needs to take into account the occurrence of accidents such as fire, corrosion, and heavy pressure. This series of hard-core test articles is very useful for users, and it also reminds the general public well. Hardware wallet.

Article 2, 2 Learn about Bitcoin's new technology innovation

An article from Bitcoin Asset Stack from Digital Asset Research Director of Technology Research Lucas Nuzzi, the author introduces Bitcoin's technological innovation in 4 aspects: Layer 2, smart contracts, mining protocols, and privacy technologies.

Article link:

Free and easy comments: The content of the smart contract part can be said to be very exciting. It turns out that Bitcoin may also implement applications such as decentralized lending (DeFi). Bitcoin is not static.

2, 3 technical dry goods | In-depth understanding of Zcash's zero-knowledge proof system

Zero-knowledge proof algorithm researcher Jiang Xiaobai wrote a scientific article on Zcash zero-knowledge proof system. This article introduces the protocol details of the Zcash Sapling version.

Article link:

Free and easy comments: As the first cryptocurrency to adopt a zero-knowledge proof protocol, Zcash's technological development has great reference significance.

Eruption and Demolition of Ethereum Difficulty Bombs

An article on popular Ethereum difficulty bombs from Thomas Jay Rush. Ethereum enthusiasts Zeng Yi & A Jian translated the Chinese version. The author showed the specific removal principle of the difficulty bomb with codes and specific formulas.

Article link:

Free and easy comments: Around January 6 next year, Ethereum will remove the third difficulty bomb through a hard fork. This article is helpful for those who want to understand the difficulty bomb and the principle of removal.

2. 50,000-word long text explaining the latest development of Ethereum

An article from the Ethereum Foundation's official website, completed by 15 development teams, subdividing Ethereum R & D on different clients, Ewasm, formal verification, Remix, Plasma, ETH 2.0, Solidity, status channel, ZoKrates, etc. The work is summarized.

Article link:

Easy and Easy Comment: It is undeniable that the Ethereum developer ecosystem is very powerful. The article seems to show a muscle meaning. However, the Ethereum Foundation also faces the question of whether to retain developers (rumored that the Foundation will stop next year Funding for most development teams), after all, it does not have much available funds, and other rich public chain project parties have the opportunity.

3. Technical Progress of Mainstream Blockchain Projects: Ethereum Successfully Completes Istanbul Upgrade

3, 1 Ethereum successfully completed Istanbul upgrade

On December 8, Beijing time, Ethereum successfully activated the Istanbul upgrade at a block height of 9,069,000. This upgrade involves EIP-152, EIP-1108, EIP-1344, EIP-1884, EIP-2028, and EIP-2200. Improvement proposals.


Free and easy comments: For some layer 2 solutions based on SNARKs and STARKs, such as zk-rollup, this upgrade is of great significance. It can improve the throughput of the solution. However, Ethereum itself has the third difficulty bomb The number of triggers will increase the interval between each 100,000 blocks, which also affects the use of the network. According to the plan, Ethereum will perform a hard fork around January 6 to remove the difficulty bomb. .

3, 2 Bitcoin development to update

This week, Bitcoin Optech introduced some discussions of last week's Bitcoin developers about Schnorr signatures and Taproot proposals, and noted the latest updates previously called OP_CHECKOUTPUTSHASHVERIFY and OP_SECURETHEBAG proposals. In addition, the article summarizes popular Bitcoin infrastructure projects Some significant changes. Free and easy comments: As the earliest and most popular cryptocurrency, its developers will inevitably be very cautious. Therefore, large-scale upgrade solutions need to be implemented after a lot of discussion and improvement. Apply Schnorr signature and Taproot proposal? First look at the next version 0.20.0.

3, 3, how is the development of Ethereum 2.0? Take a look at the latest progress

This is the fifth weekly report on the development progress of Ethereum 2.0 released by the Ethereum Foundation. Last week, it is worth noting that the most popular Ethereum browser Etherscan also launched the Ethereum 2.0 service: .

In addition, last week the developers held the first Eth2 networking conference call, which provided an overall update and review of the work performed by many related teams and individuals.


And most importantly , the developer also announced the standard specifications and implementation of the BLS signature , and after the Runtime Verification team completed the formal verification and analysis of the Ethereum 2.0 mortgage contract bytecode (expected to be published at the end of this month), the mortgage contract It can be officially deployed.

Free and Easy Comment: Phase 0 of Ethereum 2.0 is getting closer and closer. What will happen then? This will be very worth looking forward to.

This week's exciting content is here, see you next week ~