With the introduction of Alaya AI’s autolabelling system, we’re excited to announce that Alaya AI’s platform structure can even be upgraded to supply three interoperable layers for Web3-native AI knowledge autolabelling assist. These three layers will kind an information optimisation community that provides real-time suggestions to particular person knowledge modules and AI functions to assist reinforcement studying and fine-tuning for our autolabelling fashions.
The up to date platform construction will include the Interplay Layer, the Optimisation Layer, and the Clever Modelling Layer (IML). These three layers are designed to be composable and interoperable for optimum compatibility with exterior APIs and dApp integration for Web3 companions.
A mix of evolutionary computation and RLHF/HITL iteration will present Alaya AI with a definitive technological edge over parallel Web3 opponents and considerably broaden our potential for cross-platform collaboration within the Web3 AI business.
The Interplay Layer
The Interplay Layer is the person frontend of our platform and connects knowledge, communities and AI functions via a easy gamified interface accessible via our Google Play and browser dApp. Customers can simply entry our platform via both e-mail verification or pockets connection and instantly start contributing AI coaching knowledge and earn a mixture of assorted token + NFT rewards. The Interplay Layer can even function some extent of entry for service integration with different Web3 companions and allow customized token rewards via customized knowledge swimming pools.
The Optimisation Layer
The Optimisation Layer gives focused knowledge sampling and automatic knowledge preprocessing via Gaussian approximation and particle swarm optimisation (PSO). This layer gives essential optimisation algorithms and verification fashions for knowledge high quality assurance and might be continually up to date in keeping with suggestions. Different key inside capabilities of the Optimisation Layer embrace person labelling, job distribution, knowledge verification and reward calculation, all of which offer obligatory autolabelling mannequin parameters for the Clever Modelling Layer.
The Clever Modelling Layer (IML)
The Clever Modelling Layer (IML) allows clever, dynamic autolabelling AI fashions via a mixture of information accumulation, evolutionary computation and RLHF/HITL iteration. The IML consists of separate autolabelling modules distinguished by knowledge class and space of experience.
The IML might be supported by knowledge suggestions from the Interplay and Optimisation layers and up to date periodically via RLHF/HITL fine-tuning. This may allow dynamic mannequin changes to fulfill various knowledge necessities, e.g., accuracy tolerance interval vs pattern dimension.
Autolabelling AI Mannequin Staking Swimming pools
AI mannequin staking swimming pools can even turn into a key function of the IML and kind an important ingredient of Alaya AI’s DAO governance. Every knowledge autolabelling module on the IML is supported by a staking pool. Customers can stake $AGT to incentivise improvement for every mannequin and obtain a portion of the income generated by the autolabelling mannequin.
$AGT staking will enable our core customers to take part in key governance choices and allow customers to vote on subsequent autolabelling mannequin developments.
