Better-Pro AI Architecture
The network nodes of the Better-Pro AI intelligent quantitative trading system can be large computing power nodes, such as large computing pools, or medium-sized computing power nodes or use Azure or Alibaba Cloud to provide computing power. Better-Pro AI intelligent quantitative trading The cloud computing power and edge computing power obtained by the system will ensure the stability of the artificial intelligence operating environment.
After investors deposit USDT into the Better-Pro AI intelligent quantitative trading system, AI intelligence will search, identify, stratify, manage, and save the amount of USDT, and then find the corresponding ID and computing power node through the SAMSUNG network , search for a trading model suitable for this batch of USDT on the chain, and allocate the computing power required by the model to the corresponding ID to conduct quantitative transactions.
Big data+AI
The Better-Pro AI intelligent quantitative trading system contains a distributed high-performance computing network, which is essentially a big data + AI infrastructure. It has the following characteristics:
The bottom layer adopts the Matrix platform, adopts the software architecture of Topic subscription + Event driver + processor, adopts a high-performance non-6. network consensus mechanism, and adopts the AI-D-POS consensus mechanism to combine user rights and user network contributions. Assessment uses hierarchical pipeline processing, storage, consensus, and verification pipeline hierarchical processing to improve the system scalability of the entire network.
Multi-chain consists of main chain + working chain. The main chain contains the schema definitions of all work chains, and the number of work chains is not limited; the work chain is composed of sharding blockchain and supports unlimited sharding.
Block data is defined based on sema, block size supports flexible definition, and block data supports compressed storage;
Better-Pro AI’s communication transmission uses self-encoding compression protocol or encrypted transmission protocol to reduce network bandwidth, and the codec protocols of different message packets on the same link can be negotiated to improve transmission security;
Three-dimensional network: Based on the peer-to-peer P2P network, a cascading network is established to accelerate message dissemination; SN nodes are geographically dispersed and the network topology is dynamically adjusted to improve the robustness of the entire network and resist DDOS attacks. Reduce security risks;
Network architecture
The mainstream solutions to solve the scalability of blockchain are mainly divided into two types: vertical expansion and horizontal expansion. Typical vertical expansion solutions include: optimizing consensus algorithms, large block storage, and improving system performance, such as: block expansion, EOS DPOS, VRF+POS+PBFT; typical horizontal expansion solutions include: Lightning Network, multi-chain Architecture and sharding architecture are all designed to solve the scalability and scalability issues of the blockchain.
Blockchain 4.0 requires the construction of a high-performance distributed business system on a large-scale decentralized network, focusing on commercial implementation. Any single technical point improvement cannot completely solve the overall scalability problem. The overall architecture design of any blockchain needs to consider four elements: P2P network architecture, underlying storage architecture, computing/consensus architecture, and business process design. In addition, in the blockchain world, only two of the three design attributes of decentralization, scalability, and security can be chosen at most. All scalability designs are based on this principle.
The Better-Pro AI intelligent quantitative trading system adopts a multi-chain architecture, which is conducive to improving the scalability of the entire network performance. It uses the isolation of multi-chains to safely isolate DAPP data. With the parallel execution of multiple chains, the performance of the system improves. Can achieve linear growth; however, a single chain under a multi-chain architecture still has a performance bottleneck upper limit, so the technical architecture of sharding is conducive to further improving the performance of a single blockchain, allowing for further horizontal expansion; introducing between multi-chains The cross-chain router and cross-chain relay layer support the relay of homogeneous sub-chains and heterogeneous side chains. Through anchoring between sub-chains/side chains and the main chain, they can connect multiple chains and establish unlimited Scalable blockchain architecture.
Sharding architecture
The purpose of Better-Pro AI intelligent quantitative trading system using sharding is mainly to improve the performance of a single blockchain and the scalability of the entire network. Each shard processes different subsets of transactions in parallel and reaches an on-chip consensus so that transactions can be executed in parallel.
The complete sharding architecture includes three parts:
Network sharding, computing sharding (transaction/consensus), storage sharding (state sharding).
First of all, the Better-Pro AI intelligent quantitative trading system divides the nodes in the network into different virtual shards (that is, grouping) according to the sharding rules; here it is necessary to ensure the connectivity of the network within the shards and the connectivity within the shards. Network connectivity between nodes and the main network; transactions within shards are only propagated within shards to reduce bandwidth consumption;
Secondly, all nodes that need to participate in the consensus need to pay a deposit and register in the smart contract; we can shard the consensus nodes according to the sharding rules, and transactions that occur in different shards are packaged and verified by the consensus nodes of the corresponding shards; as Different shard transactions are verified in parallel between different shards, and the consensus performance can be linearly expanded;
If all nodes in the entire network store all the ledger data, as the network performance increases, assuming that the entire network performance is 1 million TPS, and the ledger data space stored by a single node is >20TB per day, it is definitely not feasible for a single node to store the entire ledger data. , the Better-Pro AI intelligent quantitative trading system separates the storage layer and the computing layer. The ledger storage is stored by the Ledger File System, and the Better-Pro AI intelligent quantitative trading system network jointly maintains the ledger data.
Programmable virtual sharding vShard
Each node of the P2P network has an independent PeerID, forming an unstructured and structured P2P network with each other;
According to different business needs, based on the P2P physical network, network nodes are virtual sharded through smart contracts; each node calculates the vShard ID of the node based on the programmable smart contract and belongs to the same vShard ID nodes form logically independent network shards;
Each network shard physically originates from a unified P2P network; it is logically isolated, and its logical independence satisfies the scalability design of the network; each independent P2P network node is logically can belong to multiple virtual shards;
Network connections need to be established between shard nodes to form a regional autonomous sharding network;
Sharding is dynamic and elastic, and the network can flexibly add sharding;
The mainnet can still establish unstructured and structured P2P networks.
Calculate shards
Consensus node: divided into global consensus node and intra-shard consensus node; global consensus node is responsible for main network global consensus; intra-shard consensus node is responsible for consensus within the shard;
Consensus can adopt a plug-in mechanism, and different shards can support different consensus mechanisms;
Distributed random generation protocol: The random protocol needs to be unbiased, unpredictable, third-party verifiable and scalable;
The main sharding consensus node is sharded according to the VRF algorithm and is randomly responsible for the consensus of different shards; the consensus nodes of different shardings are responsible for the transactions of different shardings;
Witness node selection: Randomly select from the sharded consensus node pool as the node that actually participates in the consensus;
Consensus nodes are elected periodically to avoid malicious attacks;
The consensus nodes of each shard can flexibly increase the number of shard consensus nodes;
Transaction sharding
Better-Pro AI intelligent quantitative transactions are divided into intra-shard transactions and cross-shard transactions: the shard transaction address is the transaction address + vShard ID; if the sender address and receiver address of the transaction belong to the same shard, it is intra-shard Transaction; if the sender address and receiver address of the transaction belong to different shards, it is a cross-shard transaction;
Intra-shard transactions: Transactions are only propagated within vShard shards and are packaged and verified by the shard consensus nodes;
Cross-shard transactions: packaged and verified by the main shard consensus node;
Data crawler
The operation of the Better-Pro AI intelligent quantitative trading system is based on data collection, data analysis, data modeling and other capabilities. The Better-Pro AI intelligent quantitative trading system collects data in the entire network cloud in a targeted, regular and quantitative manner through artificial intelligence, and seamlessly combines it with APIs. Connect to the AI quantification system, and then perform screening, filtering, analysis, and modeling to provide visual text to back-end personnel and provide data support for artificial intelligence.
Trading engine
In order to achieve top information flow processing capabilities and ensure accurate information arrival and error-free processing results, Better-Pro AI adopts a self-developed transaction engine system. This engine system has been tested to achieve an ultimate peak transaction processing speed of 5 million TPS. , the transaction matching efficiency is 35%-40% higher than that of the same industry, providing basic technical support for the stable and efficient operation of the platform; at the same time, the Better-Pro AI platform will integrate and optimize the cloud computing configuration of each node, so that the system can reach the top international stock futures The processing speed of the trading platform.
Large financial model
Through years of technology accumulation, we have successfully developed an LLM (Large Language Model) tailored for the financial industry: BPL-1.
This model has the following advantages:
Powerful natural language processing (NLP) capabilities: BPL-1 has excellent NLP capabilities and can understand and interpret a large amount of natural language data in the financial field, including news, social media comments, financial reports, etc. This enables Better-Pro AI to gain valuable trading insights from diverse information sources.
Deep sentiment analysis: BPL-1 can not only understand the surface meaning of text, but also perceive emotions and moods, and can better understand the emotions and psychological states of market participants and incorporate these factors into trading decisions.
Trading decision support: BPL-1 provides more comprehensive support and suggestions for trading decisions. By integrating a large amount of market information and intelligence, BPL-1 can generate more intelligent and rational quantitative trading strategies.
The technical specifications of this model are as follows:
Training parameters: 13B.
Reasoning capabilities: The BPL-1 model has powerful reasoning capabilities and can process and analyze large-scale data sets within milliseconds, providing instant market insights and responses. Its efficient computing performance ensures instant decision-making and execution in high-frequency trading environments.
Model accuracy: The model has been extensively tested and verified, showing a prediction accuracy of up to 79.2%, and the prediction accuracy of market fluctuations in simulated trading scenarios is extremely high.
Data processing capacity: BPL-1 can process more than 10TB of financial data every day, supporting instant analysis and understanding of global market events.
Learning and adaptability: The model can self-learn and adapt to market changes, and continuously updates its algorithm to reflect the latest market dynamics and trends.
Decentralized training and operation
In order to ensure the inclusive financial characteristics of the BPL-1 model and ensure that BPL-1 artificial intelligence is not manipulated by any institution, including the creator of BPL-1, Better-Pro AI established a set of financial institutions based on BPL-1 at the beginning. A decentralized training system for the chain.
We know that artificial intelligence models require a large amount of GPU computing resources in both the training and service phases. However, currently GPU computing resources are very expensive, limiting innovation in most projects. Based on our platform, any institution or individual can provide their idle or excess GPU resources to train BPL-1 while ensuring privacy and security. The latest BPL-1 model after training is completed also runs on these GPUs provided by the user. The GPU resources of the entire network will provide computing power support at any time. These resources will be used by the financial public chain platform according to its economic model. There is no need to worry about insufficient resources due to rapid user growth, nor to worry about excessive resources and high fees.
In this system, the hardware resource providers are called miners. Miners share their GPU resources with the entire network through the distributed computing terminal of the financial public chain, and receive platform currency rewards from Better-Pro AI to ultimately complete the economic closed loop.
We can conclude that through decentralized training and operation, BPL-1 will become the fairest financial AI with everyone for one and one for everyone. It will also become the largest gathering of computing resources and the fastest AI evolution. Public AI.
Last updated