Post Idea: Leveraging Blockchain for Greater Accessibility of Machine Learning Models

Blockchain AI

To quote the brief summary I had made of it:
This Stanford article highlights the fact that the processes in machine learning are frequently cost-inefficient and difficult to access for many. Datasets are frequently privately-owned and expensive to create. Therefore, it’s all about using blockchain to democratize AI.

Blockchain offers trust and security. Through smart contracts, it also offers reliability in execution of code and transparency. Smart contracts are specifically what is used for contribution of data. Hosting an ML model requires a one-time fee for uploading and a small fee for anyone uploading data. The authors propose three methods for incentivizing people to upload “good” data: gamified (rewarding based on validation and acknowledgement from other contributors), prediction market-based (reward based on how the data submitted affects the model’s accuracy), and ongoing self-assessment (reward based on mutual validation and mutual payment for good data).

Right now, this framework is most useful for smaller models, and research in scaling still needs to be done. Overall, however, a promising area of research!

4 Likes

This is a great idea. I was thinking about putting something together on the role of machine learning/AI in cryptocurrency transactions and oracles and this is really helpful.

3 Likes

This seems like a great paper to have on the forum! Possible collaboration between @Gearlad and @jasonanastas?

3 Likes

I had actually moved this paper to drafting earlier today, as I am working on it. It is not that long of a paper, but the subject itself warrants more collaboration.

Research Summary: Leveraging Blockchain for Greater Accessibility of Machine Learning Models · Issue #138 · smartcontractresearchforum/research_content (github.com)

5 Likes

I’m glad you’re on it! Looking forward to it hitting the forum in the near future.

3 Likes

@Gearlad Great idea! Decentralized the datasets and bring a more efficient machine model is definitely the future trend of AI. It’s concepts such as federated learning in this paper “When Federated Learning Meets Blockchain: A
New Distributed Learning Paradigm
” written by a couple of IEEE members.

5 Likes

This is great, thanks for the reference - we’ll add it to our list of papers for the AIoT Blockchain project.

5 Likes