Platform Overview

An overview of the Spectral Network, key features and functionalities.

Spectral streamlines an innovative use case: inference feeds for smart contracts. While existing oracle networks cater to data feeds, most notably for price data, Spectral addresses a significant gap by enabling an unprecedented feed of high-quality inferences. This extension broadens the scope and the potential market of smart contracts in general, opening new horizons for decentralized applications in Web3.

The Network offers bounties, revenue sharing, and perpetual competitions to incentivize independent ML modelers to create high-quality models and inference feeds onchain for Creators and Consumers who post bounties, set benchmarks, and subscribe to constantly updating feeds. Validators receive fees by ensuring the process remains fair.

This dynamic is called the Inference Economy. It will bridge the gap between ML, AI, and the blockchain. Our mission is to empower and reward the intellectual contributions of data scientists and ML engineers (who we call “Modelers”) for solving real-world data science challenges while providing third-party consumers with high-quality, tamper-proof inferences directly on-chain.

Model Quality and Security

Our goal is to enable Modelers to build, validate, and deploy the most accurate, highest quality ML models. Each model must adhere to a stringent process which includes model commitment and verification through zero-knowledge machine learning (zkML). By allowing modelers to mathematically verify that their models produced an inference without revealing its proprietary methodology, zkML certifies the integrity and performance of the models and preserves the intellectual property rights of the Modelers.

Validation and verification are built into the platform. By enabling Modelers to share their innovations without forfeiting their competitive advantage, Spectral will foster a nurturing community environment for intellectual growth and contribution. Spectral is also implementing an automated and provably fair mechanism (performed by whom we call “Validators”) to evaluate all submitted models against challenge-specific performance benchmarks. This ensures that only models which meet or exceed benchmarks are considered for consumption. As the platform evolves, it will integrate a network of Validators, effectively decentralizing the validation process to become even more secure and transparent.

Modeler Incentives and Obligations

Spectral incentivizes Modelers through a repeatable, ongoing 'model-to-earn' system, made possible in part by our Modelers’ ability to retain ownership of their models and to keep their model details confidential. This system rewards Modelers for addressing data science problems with their expertise, to monetize their machine-learning skills through transparent and adaptive payouts that are directly correlated with the consumption of their models. For Modelers, meeting and maintaining performance benchmarks is crucial, as it determines their proportional share of the challenge bounty.

To earn a bounty, a Modeler needs to be active “online”—in this context, it means a model must be ready to respond to consumption calls. Earnings are unlocked only when a model's inferences are consumed in real-time. This setup encourages Modelers to ensure their models are continuously optimized and available, as the frequency of inferences being consumed directly impacts the portion of the bounty unlocked.

The incentive structure is also designed to reward ongoing participation. Even after the initial bounty for a challenge is fully distributed, Modelers continue to earn a share of the live consumption fees. The distribution of these fees is dynamic, depending on the number of competing Modelers who are online and responding to consumption calls. Fewer Modelers online translates to a larger share of the rewards for those present, while more Modelers online means the rewards are distributed more broadly. For instance, if a challenge offers an 80% revenue share and is limited to the top 10 Modelers, a share could range from the entirety of the 80% revenue, if there’s a single respondent, to a 10% share (of the 80%), if all top Modelers are active. This mechanism ensures a fair and balanced distribution of rewards, reflecting each Modeler's contribution to the platform's ecosystem.

Challenges on Spectral are continuously open, allowing new Modelers to submit models to outperform existing solutions. Successful Modelers can earn a share of ongoing payouts, encouraging constant improvement and innovation. The perpetual challenge framework ensures that the platform remains at the cutting edge of ML solutions, as it incentivizes Modelers to continually refine and enhance their models to stay ahead.

The Spectral platform was designed to host a diverse array of data science challenges aligned with real-world problems. In the first phase, the platform will focus on Web3-native challenges, opening up possibilities for practical problem-solving and secure ML applications. With Web3 challenges, the ground truth for training and testing models is retrieved directly on-chain and facilitated by Spectral’s data wrapper, custom tooling that facilitates the pre-processing of feature engineering.

Web3 and Beyond

The Spectral platform is designed for data scientists, and although it relies on Web3 technologies, throughout our design process we’ve tried to make using these systems as intuitive and user-friendly as possible. Specifically, we use wallet interactions, multi-signature wallets, and on-chain payouts. These abstractions are essential to ensuring Modelers of various backgrounds can leverage these advanced systems without needing to navigate their complexities. By simplifying the user experience, Spectral allows Modelers to concentrate on their core expertise—developing potent and innovative models—while the platform handles the underlying blockchain mechanics. This approach not only broadens accessibility, inviting a diverse range of talents to participate, but also accelerates the adoption of high-quality, tamper-proof, and privacy-preserving crowd-sourced intelligence.

The Spectral platform represents the convergence of cutting-edge technologies, creating a unique synergy between ML and blockchain technologies. This integration is pivotal for the future of predictive analytics, offering a decentralized network that not only solves complex predictive problems but does so with an unprecedented level of security and privacy. Spectral is not just a platform; it is a new standard for applying competitive machine learning in the modern world.