Spectral
Ask or search…
K

FAQs

Welcome to the Spectral Platform FAQ section! Here, we address common questions and concerns to help you make the most of your learning journey. Whether you're new to data science or an experienced practitioner, we're here to provide guidance and support.

What is the Spectral Platform?

Spectral is a decentralized on-chain machine learning competition platform that harnesses the power of Zero-Knowledge Machine Learning (ZKML). It operates with four key roles: Creator, Modeler, Validator, and Consumer.

How does Spectral differentiate from the other data science crowdsourcing platforms?

  • Modelers enjoy fee-sharing for each Consumer request while never having to give away their ML models
  • Creators create any type of ML competitions/tournaments to help solve problems in any relevant domains
  • Validators allow for maximal decentralization and unprecedented transparency
  • Consumers receive high-quality ML inferences that are verifiable via ZKML
  • Many on-chain use cases such as for web3 credit scoring, trading, MEV, recommender engines, risk forecasting, etc.

What is zkML?

Zero-Knowledge Machine Learning (zkML) is an innovative approach that combines the privacy-preserving aspects of zero-knowledge proofs with the predictive power of machine learning. In essence, zkML enables the training and utilization of machine learning models in a way that the underlying data or the model's specifics are never fully revealed or compromised. This means a model can be used to make predictions or analyses without ever exposing the sensitive data it was trained on, nor the details of the model itself.

How does Spectral use zkML?

Spectral uses zero-knowledge machine learning (zkML) to train machine learning models without exposing the underlying data. This technique allows data to remain encrypted and secure throughout the process.
Our CLI and SDK abstract this process away for the user.

What are the advantages of using Spectral as a Modeler?

Spectral empowers modelers to maintain the privacy of their models through zero-knowledge machine learning (zkML). Additionally, it enables modelers to monetize their models either by securing bounties or through revenue sharing when consumers use their models. This approach offers a unique balance of privacy and profitability for modelers.

Is Spectral compatible with existing machine learning frameworks?

Yes, Spectral is designed to be compatible with popular machine learning frameworks. This ensures ease of integration into existing workflows and allows users to leverage familiar tools and libraries.
Each challenge may have specific model requirements.
Popular choices include TensorFlow, PyTorch, Scikit-learn, and Keras.

Why do I need to use a CLI?

The Spectral CLI, as part of our Modeler SDK, is an essential tool for model submission, model commitment, and accessing various functionalities, ensuring seamless interaction with the platform.

Why do I need to create a multi sig?

Creating a multi-sig wallet is crucial for enhancing the security and integrity of transactions on the Spectral platform, enabling a decentralized and transparent validation process.

How do I become a top Modeler?

Consistently submit high-quality ML models that meet or exceed the challenge-specific performance benchmarks. Active participation, maintaining high model performance, and engaging in the ongoing 'model-to-earn' system are key.

What is Rolling Validation?

Rolling Validation refers to the continuous process of evaluating submitted models against the set performance benchmarks. It ensures that only the models which continually meet or surpass these benchmarks remain active and eligible for rewards on the platform.

Do I need to generate a ZK proof by myself?

As a Modeler, you're responsible for generating zero-knowledge (zk) proofs for your models. These proofs validate that your model produced an inference without revealing its proprietary aspects, thus preserving your intellectual property rights.
Our CLI and SDK abstract this process away for the user.

What is the Consumption Window?

The Consumption Window is a designated time period during which your model's inferences are consumed in real time. Earnings are unlocked only when inferences from your model are used during this window, incentivizing continuous optimization and availability of your model.

Can we form teams for the challenges?

Yes, teams are welcome! However, model submissions must be made from a single wallet address.

Are there any rules on data privacy and model ethics?

Yes, participants must adhere to data privacy regulations and ethical modeling practices. Models should not incorporate or infer sensitive personal information.

What happens if two teams submit a model with the same performance score?

In case of a tie, the winner may be determined by the earliest submission time, or additional criteria such as model complexity and resource efficiency.

Can I update my model after submission?

Yes, you can update your model after submission. The platform encourages continual refinement and enhancement of models. However, each updated model must be re-submitted and will go through the validation process to ensure it meets the necessary performance benchmarks.

Is there any support available if I encounter issues during the challenge?

Yes, we provide support through our Discord and Slack, specifically in our or direct support from our team for technical or competition-related queries.