Friends who follow artificial intelligence or technology news must be familiar with Google's DeepMind. In March 2016, AlphaGo, a program developed by DeepMind, beat the Korean Go champion Lee Se-dol 4:1, creating a sensation and becoming a milestone event in the field of artificial intelligence in recent years.

Recently, a team of former DeepMind artificial intelligence researchers has also brought AI technology to the blockchain field.

On November 16, Kosen Labs announced the completion of a US$5 million seed round of financing, led by a16z and participated by Framework Ventures and other strategic partners. Founded by ex-Google DeepMind AI researchers and engineers Miljan Martic and Peter Toth, the Kosen Labs team has extensive experience in designing and implementing cutting-edge solutions to machine learning and numerical optimization problems. The mission of Kosen Labs is toBring the advantages of AI and advanced optimization to decentralized applications such as DeFi, while lowering the threshold for user participation, designed to allow everyone to participate in the decentralized economy.

As a first step in this mission,Kosen Labs will focus on providing best-in-class active liquidity provision (LP) solutions for next-generation automated market makers (AMMs) such as Uniswap v3 and Sushiswap Trident.

In this regard, Kosen Labs built a proprietary AI research stack using historical on-chain data, data from centralized exchanges, and a custom strategy backtesting engine, and developed the Kosen Labs v1 strategy, which uses machine learning in Uniswap v3 It is a full-featured solution for automated liquidity management on .

The v1 strategy combines machine learning models with constrained optimization. It includes 3 main components:

• Machine Learning Price Fluctuation Prediction Engine, to predict the price distribution over the desired time period, which in turn enables users to use non-uniform liquidity allocation strategies, which are more liquid and capital efficient.

• Constrained Optimization Algorithm, to make forecasts and output decisions about the number of positions and their parameters to maximize capital efficiency and liquidity coverage under the provided distribution.

• Position Management System, to track all current open positions and decide when adjustments are required.

The v1 strategy is ready to be plugged into L2, and because the solution is composable and the machine learning model can be easily adapted to different situations, the v1 strategy can trade off between higher gas usage and higher capital efficiency at any time .

Kosen Labs first chose the design of the solution in DeFi, especially in the field of active liquidity provision, because:

First, DeFi is programmable, with open APIs and vast amounts of on-chain data accessible to anyone, which is where machine learning (ML) and advanced optimization methods can come into play.

Second, next-generation AMMs (such as Uniswap v3) provide users with a better experience, with higher liquidity and lower slippage, but also with higher technical complexity. LPs need to consider many things such as centralized liquidity, asset portfolio, order types, rebalancing, and Gas fees. It is a complex control and optimization problem that requires reliable machine learning predictions to support continuous policy re-evaluation and intelligent updates of positions and parameters on a regular basis.

Finally, with Uniswap v3 and Sushiswap deploying L2 solutions, and Ethereum moving towards PoS implementation, AMM transaction gas fees will drop significantly in the next 6-12 months. If successful LPs want to remain competitive, they will be forced to use some form of advanced machine learning and optimization techniques, which will shift AMM versions of market makers to efficient algorithmic and high-frequency trading strategies.

The v1 strategy outlined above is just the first step in Kosen Labs' research agenda. Kosen Labs will plan to further develop more complex volatility models, optimally constrained optimization and better position management algorithms, and eventually form the deployment of a full-fledged end-to-end reinforcement learning system capable of self-selecting all parameters and steps, Also take into account the conditions provided by the user.

This strategy will first be applied to Uniswap v3, and then it will consider expanding to Sushiswap and other potential AMMs, as well as releasing vaults on L2 scaling solutions. It will also actively cooperate with other projects in the DeFi field that seek to optimize yield generation strategies.

The ultimate goal of Kosen Labs is to connect advanced AI and machine learning with the Web 3 world of the decentralized economy and create greater value at the intersection of these two fields.

In terms of tokens and governance, the native token of Kosen Protocol is KOSN, and the protocol will be managed and operated by Kosen DAO in the future.