Trusted machine learning on blockchain

February 12, 2021
Trusted machine learning on blockchain

IoTeX and Scaleout announce partnership to enable trusted end-to-end machine learning and decentralised machine learning models

Machine learning is about learning from data. In general, more high quality data gives better machine learning models. With insufficient data, it is not possible to train any useful machine learning models at all. Today, most useful and valuable machine learning models are created by one entity that controls the entire AI value chain from data control to integration with product and services.

Scaleout and IoTeX are now starting a joint project to unbundle the AI value chain and bring collaborative trusted machine learning to the IoTeX platform. The end goal is a fully trusted end-to-end solution for machine learning, from collaborative model training to smart contract based model governance and on-chain serving of machine learning predictions.

This will be a long term effort, and work will proceed in several phases. The first phase will develop a solution to jointly train machine learning models on trusted and verifiable data from IoTeX Pebble devices, and then provide a solution that serves predictions from these models on-chain and off-chain to end users.

The first phase will also enable the creation of the first collaborative machine learning networks on a blockchain, where machine learning models can be co-owned by those participating in the creation and training of a model. Collaborative machine learning has the potential to change the current paradigm around data control and data generation and enable market leading machine learning models that are not owned by one single entity, but by all those who have participated in training the model. Imagine contributing to, and therefore owning a stake in a customer flow model for local shops, or a personalised search engine that respects personal privacy!

Figure 1: First phase of the IoTeX and Scaleout solution where trusted data is made available to train ML models, and smart contracts are enabled to make call models and serve predictions on or off-chain.

The long term objective of the IoTeX and Scaleout collaboration is to bring machine learning capabilities (tools and APIs) to the IoTeX platform and to enable fully trusted end-to-end machine learning pipelines.

Development roadmap

The first phase will focus on the ends of a typical machine learning pipeline — data in and predictions out. Pebble trackers and the IoTeX platform enable the use of trusted data in, and this project will develop a secure and trusted way to train, govern and access machine learning models.

Introducing machine learning collaboration networks

We will also enable collaboration between peers and devices that generate trusted data for training machine learning models. This will require infrastructure and tools to access the data and train the machine learning models, and smart contracts to govern control over these collaborative models. Stay tuned for examples and boilerplate code to launch your own experiments!

Off-chain infrastructure to build, train, validate and service machine learning models

Machine learning is generally an expensive task in computational terms and uses a lot of compute, storage and network resources. It is not feasible (or even wanted) to do everything on-chain, so bridges between on- and off-chain machine learning resources will be developed.

Integrate models into production systems

A machine learning model is only any good if you can put it to use. There will be methods to access machine learning models on-chain and off-chain.

Summary of deliverables — Toolkit to launch machine learning collaborative networks
  • Tools to build and train machine learning models using on-chain IoTeX trusted data
  • Tools to deploy and run machine learning models built on IoTeX data
  • Tools to deploy and run service endpoints with access and compliance rules such as access, distribution and model inference rate-limit policies
Deliverable 2 — Framework to integrate models in production systems
  • Tools to serve machine learning models as API endpoints
  • Frameworks to provide models with attached incentive mechanisms
  • Services to engage with published model service registries and model providers
  • Tools to govern the machine learning pipelines and machine learning alliances on the IoTeX platform
Learn more in this discussion between Daniel Zakrisson and Larry Pang and article below.

Medium article by Iotex.

. . .

About Scaleout

Scaleout is a team of data scientists, machine learning engineers, software engineers, and entrepreneurs. Experienced from both industry and academic research in AI, cloud and fog computing, and scientific computing from top-ranked Uppsala University in Sweden. Collectively the team has authored 100+ peer-reviewed articles in AI/ML, scientific computing, and systems biology.

Scaleout is pioneering federated learning to overcome the data-sharing challenge, building collective intelligence from distributed data at scale, while preserving data privacy and security. The Scaleout platform enables federated machine learning, data partnerships, and joint machine learning ventures. It is built for creating commercial machine learning alliances where data owners can build strong machine learning models together.

We’ve worked with some of the largest and most reputable organisations in the world, including AstraZeneca, SAAB Defence Systems, Swedish National Space Agency, Autodesk, Raysearch Laboratories, GE Healthcare.

Learn more: Website | Twitter | LinkedIn

About IoTeX

Founded as an open-source platform in 2017, IoTeX is building the Internet of Trusted Things, an open ecosystem where all “things” — humans, machines, businesses, and DApps — can interact with trust and privacy. Backed by a global team of 30+ top research scientists and engineers, IoTeX combines blockchain, secure hardware, and confidential computing to enable next-gen IoT devices, networks, and economies. IoTeX will empower the future decentralized economy by “connecting the physical world, block by block”.

Get in touch to learn more!

Enter your email below

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.