- Developed by: Fredrik Liu, Lupeng Yang, Siyu Chen
Let me make this clear first, machines do not learn and smart contracts are not smart. Why? Because if we slightly distort the inputs, the output is very likely to become completely wrong. A machine learning algorithm, if trained by 'looking' at an object straight, will fail to recognize the object if rotated unless it was also trained to recognition rotation. (Though Fred is working on 3D equivariant neural networks that can handle translation, rotation, and inversion under Euclidean space). For smart contracts, which are hard coded to do only one thing, has proven to be neither smart nor even enforceable for the case of the DAO. But Blockchain is still an truly innovative approach to governance for networks and machines.
Our vision at DeepVerse is to integrate machine learning and blockchain to do various useful things without being explicited programmed to and yet remain open, transparent, and decentralized.
+ Bridging the digital world and the real world.- Machine learning backends
- Model hosted decentralized
- Decentralized storage
- API calls via Oracles
- Smart contract logic
Last-mile problem! What blockchain can not do!
Blockchain is good at sharing digital records and assets, where it can replace the trust between players or a central authority to verify. We have seen people claims to use blockchain for tracebility and supply chain management. The digital records may be immutable and verifiable, but how does someone know which digital record is attched to an real-life object or human. To link an entry on the blockchain, we need a physical identifier whether a small chip or QR code to links to its digital record. And this is blockchain falls down.
We are using machine learning algorithms to establish the link and solve the last-mile problem, so it can be used not only to ensure data integrity but also to give individuals control over how their data is used (for academic research, for COVID contact tracing, for drug deveopment, etc.)
verfication for objects on-chain and off-chain without a centralized human intervention or operation
-
Deepverse for product tracing As we all know, the tamper-proof properties of blockchain are most suitable for product tracking. However, traditional schemes are easy to be deceived because there is no mechanism to prevent people from uploading fake data to. Now, with the Deepverse's solution, you can customize a machine learning model for the product you want to track (for example, let’s say an apple recognition model), then verify if each record to be uploaded to the blockchain really means an apple.
-
Deepverse for NFTs