Thread by @DCbuild3r

An introduction to Zero-knowledge Machine Learning (ZKML). Creating ZK proofs of ML models allows us to ensure that a given output O was created by model M using input I.

Link: worldcoin.org/blog/engineeri…

1/

For example, we could prove that some text was produced by a large language model with a specific prompt and that an image was created by a given text-to-image model with a given prompt. Currently, the size of models that we can create ZK proofs of is relatively small.

2/

According to @ModulusLabs' new paper, it takes about 50 seconds to create a ZKP of a model with 18M parameters using the plonky2 proving system on a powerful AWS machine. Read more in their recent accompanying blog post: medium.com/@ModulusLabs/c…

3/

There are several teams working on improving ZK technology, creating optimized hardware for the operations that occur inside ZK proofs, and creating optimized implementations of these protocols for specific use cases.

4/

As these technologies mature, bigger models will be ZK-provable on less powerful machines and in a shorter period of time. We hope these advancements will allow new ZKML applications and use cases to emerge.

5/

In the context of Worldcoin World ID users would be able to self-custody their biometrics in the encrypted storage of their mobile device, download the ML model for IrisCode generation and create a zero-knowledge proof locally that their IrisCode was created successfully.

6/

They'd be able to permissionlessly insert the IrisCode into the WorldID protocol without having to visit an orb again. It could also prove that the orb is running fraud and sybil resistant checks during sign-up and show it to the user to verify that it is a secure orb.

7/

These are one of the many potential use cases which we mention in the full blog post! ZKML is a really exciting technology and we are thankful to many members of the wider community for their contributions and enthusiasm.

8/

Thank you @jasonmorton @CamutoDante @nayr_oac @realDanielShorr @daniel_d_kang @theyisun @drCathieSo_eth and everyone else that is contributing to the early days of ZKML!

9/

To learn more you can join the ZKML community (t.me/zkmlcommunity) and visit the awesome-zkml aggregator of ZKML resources (github.com/zkml-community…).