Bytom CTO Lang Yu: Blockchain Has Immense Impact On AI Edge Computing
March 29 2018 – LANG Yu, CTO of Bytom and founder of 8btc News, delivered a speech titled “Blockchain Has Immense Impact On AI Edge Computing” in the 2018 China Blockchain Technology and Application Summit Forum, talking about the combination of two leading technologies of today – blockchain and AI computing.
About cloud computing and edge computing
What is a cloud computing platform? All data, including data on face recognition and behavioral characteristics recognition, are uploaded from the terminal devices to the cloud computing platform and generate corresponding data through a series of complicated computing.
Edge computing is hot recently. For instance, Apple’s chip for face recognition, it stores face data on its own mobile phone chip instead of uploading data to the cloud. Edge computing, in my opinion, is the future trend of post-AI development.
Today I’m talking about how blockchain can be combined with edge computing and AI computing.
What is Bytom?
The word “Bytom” comes from the combination of byte and atom. Our positioning is a dedicated public-chain protocol for the asset realm, with the goal of bridging byte assets (including various tokens and digital assets) with atomic assets (substantial asset like bonds and equity). Any user or organization that needs to enable its assets flow freely are our clients. Considering the entire financial market, including the financial derivatives market, it is definitely a broad market, a trillion-dollar market. While at present, byte assets only take a very small share of the entire financial market, so we are facing a very broad market with great potential.
The architecture of Bytom
The slide above tells the architecture of Bytom, which achieves four layers of innovation. They are:
- Asset unique – realizing the bridging of byte world and atomic world;
- UTXO model – a paradigm similar to that of bitcoin and bit different from ether;
- Smart contract – a particularly designed contract system based on financial asset transaction;
- AI-hash Pow/ Tensority – an exclusive Pow-supported consensus algorithm, which removes the boundary between AI (Artificial Intelligence) and blockchain. And next we’ll mainly focus on this part.
Tensority – a tensor-calculus-based consensus algorithm
Tensority is a tensor-calculus-based consensus algorithm. We adopt the most important algorithm of matrix operation and convolution in the deep learning of AI. Before that, we need to get clear these two questions.
Why we adopt POW?
Some projects may think POW mechanism is inefficient and other proof mechanisms such as the DPOS consensus algorithm will be more efficient. So why POW? It is indeed controversial.
The first reason, we believe, is that Bytom is a protocol. The most basic requirement a public chain must meet is that it should be a non-permissive system, while other mechanisms like DPOS, PBFT, etc. are permissive algorithms. That means, you can join it only when approved by others. In this sense, Bytom is a neutral protocol, you can join in without anyone’s permission. Therefore, we believe POW is viable.
Secondly, setting hashpower barrier is the best way to establish the ecology of the public chain. As we know, the top two public chains are Bitcoin and Ethereum. It is rather difficult, maybe impossible for Ethereum to get into POS mechanism. For example, if I am not a bitcoin investor, I will definitely choose the most cost-effective blockchain. In POS, it is quite easy to copy a chain.
The third reason is that the protocol can be self-updating iteratively, and rid of single points of failure. It scarcely occurs in a POW mechanism that the entire network breaks down for a single point of failure.
Why we create a new algorithm?
Why to create a new algorithm? This is quite interesting with three reasons as well.
- SHA256, the algorithm of bitcoin, is not widely used in many devices and terminal nodes. When it comes to edging computing, we need to run nodes as many as possible in more terminal devices.
- The bitcoin mining machine is under rapid development and iteration, it is somewhat wasteful. The advent of a new generation of mining machine disables its previous generation. So we’re considering whether it can be reused, maybe combining with applications in reality business.
- The full nodes of a public chain community are a quite essential indicator to see whether the community is strong or widely accepted. It needs to attract large amount of users into the consensus participation.
The problems Tensority address
We’ve known the reason for adopting POW, as well as its existing problems, now let’s see how Tensority address these problems.
- General-purpose algorithm
First we solved the general problem of equipment. In 2017, the State Council issued the planning of a new generation of artificial intelligence, in which it was pointed out that AI would occupy a very important position in the development of the country. Tensor calculation, especially matrix multiplications, is a fundamental part of deep learning. Tensority constructs three matrixes of 256*256*256 to get the result by multiplications, and then calculates the final workload through SHA256. It is the core of our calculation. We have cooperated with the laboratories of Southwest Jiaotong University and have proved that they are randomly controllable. So we believe that it is a general-purpose algorithm. Any GPU, such as everyone’s smart phones and smart devices, can adopt this algorithm to participate in consensus and get BTM rewards.
- Reuse of devices and hashpower
Secondly, we realize the reuse of devices and hashpower, as well as AI hashpower sharing. It actually has become a popular business pattern. Take the Silicon-Valley-based Vectordash for example, it is a GPU-accelerated cloud platform which can provide GPU-powered hashpower rental service for companies or individuals. It also uses available hashpower to mine cryptocurrencies. It will make a business balance between the two.
We are in a similar situation. Our ASIC chip can either mine BTM or make AI hardware acceleration. It can be adjusted and select either task depending on different market conditions. What’s more, when the current generation of chips with higher hashpower comes out, the previous ones can turn to accelerated service. In this pattern, both the robustness of Bytom ecology and the development speed of AI chip are greatly improved.
- Well-suited operating condition
Thirdly, we create a well-suited operating condition. How to enable your devices to get into the Bytom network? AI edging computing has become the focus of major semiconductor manufacturers after the advent of FaceID. According to ARM’s financial report, the number of chips used in all kinds of smart devices is estimated to reach 100 billion by 2020. So the chip used in edge calculates will be suited in all terminal devices. In the future we will run our full nodes, probably not on a PC, nor on a server, but on all kinds of AI devices like in-home routers, cameras and mobile phones. In such a situation, Bytom is facing a market that has huge potential, where everyone can take part, and our community is getting bigger and bigger.
- One camera one vote
Our slogan is “One Camera One Vote”. Monitoring camera is the best well-suited device for Bytom. China has 176 million monitoring devices, and the number is expected to exceed the number of all mankind in five years. It is in standby mode 24/7 with access to network and storage. We’ve tested its power consumption in the case of running Bytom, its power consumption is only increased by 1%. Now there may be no such cameras equipped with chips for face recognition or pattern recognition, but we believe more cameras with AI feature will spring up along with the development iteration. In this way, we would accelerate the network building of Bytom and involve more people in Bytom’s ecology.
- The virtuous cycle between blockchain and AI
The virtuous cycle between blockchain and AI can help address the issues in the AI business mode. The cycle above goes: the more devices run Bytom – the more Bytom users will be- the higher hash rate Bytom could have – the more profitable with consensus and faster development of AI chip, the four sectors above are in synergistic and interactive relationship.
Status quo of Bytom – mainnet scheduled to go online in April
The tensority algorithm has been verified both theoretically and practically. A dedicated device would be launched partnering with Bitmain.
The mainnet is scheduled to go online in mid-April, now it is in the beta phase. All details on algorithm and whitepaper will be released with our mainnet online.
Check Bytom’s open source code on https://github.com/Bytom/bytom.