zkML on Lumia

Who this is for: issuers and developers evaluating whether verifiable machine-learning inference is relevant to their asset. This page explains the technique generally. It does not confirm a shipped zkML product on Lumia: see below.

What zkML is

zkML (zero-knowledge machine learning) is a technique for proving that a machine-learning model produced a given output from a given input, without revealing the model's weights and without requiring the verifier to re-run the inference. In an RWA context, the appeal is straightforward. If a servicing or valuation decision is made by a model, for example a risk score or a valuation estimate, zkML is a way to prove that decision was actually produced by the stated model on the stated input. Holders do not have to take the operator's word for it.

This is a general cryptographic technique used across the wider industry, not something specific to Lumia's architecture.

What is and isn't confirmed on Lumia

This is exploratory work, not a product. No zkML product, integration, or contract is live on Lumia. There is nothing here to integrate against today, no contract address, no SDK, and no timeline. Treat this page as describing a technique and a direction Lumia finds interesting.

The yield application

The specific application Lumia has explored is yield optimisation: using a model to decide how capital is allocated across strategies, and using zkML to prove that the allocation a vault actually made was the one the stated model produced from the stated inputs. Without that proof, a holder is trusting an operator's word that the model ran at all.

Lumia's public materials have used more than one name for this work. Whatever it is called, the status is the same as above: exploratory, with nothing live.