Allora is an open-source, permissionless intelligence marketplace. It brings together consumers who pay to uncover conclusions or specialized knowledge. Workers reveal these conclusions, and evaluators determine the accuracy of their work once the factual truth is exposed. Validators ensure the verification of agreement status, history, and reward distribution. With these elements, Allora can continuously improve over time and generate more accurate conclusions than the most precise participants.
The Allora adapter is currently operating on Sepolia. The Allora network is a machine learning (ML) model network incentivized to collectively optimize certain underlying objective functions. The machine intelligence network consists of multiple sub-networks, each defined by its own objective function, from predicting asset prices at a future time to constructing portfolios to optimize certain risk/reward scenarios. This creates a network with greater applicability than the individual functionalities of its sub-networks.
Machine learning and artificial intelligence (AI) play crucial roles in many instances. They utilize immense computational power and data-driven algorithms to discover insights, make predictions, and drive optimizations beyond human cognitive or current on-chain expression limits. Machine intelligence networks allow us to leverage this capability in a decentralized manner, enabling us to construct more advanced on-chain primitives.
For Allora, AI-driven decentralized finance (DeFi) is its most directly useful primitive. The network is optimized for financial applications and incentivizes network participants to contribute their asymmetric insights into financial markets (i.e., their “alpha”). Participants submit their alpha in the form of data, predictive model features, and predictions, ultimately aiming to enhance Allora.
Subsequently, an Alpha proof mechanism aggregates and rewards participants based on the usefulness of their contributions to optimizing certain objective functions (such as minimizing average absolute direction loss or maximizing the Sharpe ratio). Extra security is built into any application constructed on the network through verifiable computation. This is achieved through a new zkSNARK proof system designed to verify the outputs of tree-based ML models. Utilizing this proof system is a new tool called zkPredictor, developed in collaboration with Modulus Labs and supported by them. The result is a decentralized, self-improving intelligent network optimized for modeling financial markets.
