Tezza: AI for the Most Profitable Trading and Arbitration
Recently we've been musing about the cost of data storage in web3 social networks, video hosting, and gaming. But we haven't thought about using blockchain for secure data storage and AI training.
The Tezza team wants to create a framework for decentralized storage of data with the subsequent AI learning. Calculations of risks and profits for flash loans and farming will become the first use cases and proofs-of-concept.
For artificial intelligence to learn how to analyse objects, identify patterns and draw conclusions, it needs a large amount of qualitative data for initial learning.
In machine learning, developers give the computer data and the right answer, such as pictures of roads with instructions on how to analyse the road and its surroundings. The team then gives the AI the unsigned photos and asks it to make a guess as to whether or not there is a red traffic light. If the computer is wrong, the developers tell it so that it can adjust its future behaviour.
Machine learning produces good results, but it requires a lot of human labour. Specialists have to pick up huge sets of diverse photos, otherwise there will be problems. For example, if there are too many black cats in a set of animal photos, the AI will assume that all cats are black.
In deep learning, artificial intelligence, which is made up of neurons, learns to analyse information from labeled data too, but it can recognise much complicated patterns and provide better results. But to do it, the computer needs to provide a continuous and varied stream of data. The more the better. So there is a need for a solution to safely and continuously share data to train AI for different tasks: autopilots, pharmacy, logistics, etc.
The Tezza framework allows one to run subchains to store and transmit data securely through the blockchain protocol. In other words, nodes break down the original data set into chunks and store them in encrypted form. When a user wants to access them, they send a request to the owner. If the owner confirms it, the nodes send all the fragments to the user in a regular blockchain transaction. By using the blockchain and the Merkle tree, the recipient can ensure that no one has changed the data.
With Tezza, companies and organisations can share data sets securely and work together on machine learning. Refer to the Whitepaper for details.
Farms, Flash Loans, and Neural Networks
The Tezza team plans to collect data about profits in liquidity pools and farms on Tezos on its private chain. From this, it will teach a neural network to identify the best times to deposit and withdraw liquidity to achieve the maximum profit. They call it Farming as a Service or FaaS.
The developers have already tested it with the historical data of the DeFi dashboard YieldWatch. The annual returns in the AI case were 25% higher than average. For more details about the neural network, see the Whitepaper’s pages 6 to 8.
Flash loans are another way of making money in DeFi that can benefit from the use of artificial intelligence. When using flash loans for arbitrage, users encounter frontrunning, where the bot notices an arbitration transaction in mempool and duplicates it with a higher commission to be the first to take advantage of the arbitration opportunity and make a profit.
A trained neural network will be able to find patterns and subtle opportunities for profitable trades using flash loans. Tezza’s team trained the neural network on 10,000 examples for testing and then manually checked the execution results. About 30% of the arbitration opportunities found would produce profitable trades if executed. This is a great result for flash loans, as unprofitable trades are simply not executed and the user loses only transaction fees.
The Tezza team plans to launch FaaS and flash loans for all Tezos users, and allow other teams to deploy their subchains to share data. Users will pay commissions in TZAA tokens for working with these services.
Using artificial intelligence and neural networks to boost returns on investment is not a new idea. But Tezza could be the first project to train AI immediately on on-chain data.
How will the launch of a project that advises farms with the best APYs and picks up flash loan opportunities affect Tezos? Most likely, it will increase the activity of network participants and also help stabilise prices in liquidity pools. So, we expect Tezza to launch its beta in the first half of 2023.
Subscribe and never miss updates from the world of Tezos: