besirkurtulmus
commented
6 years ago @Swapnanandamuri Great question.
The short answer is: It's an interesting challenge.
The long answer is the following:
- Solidity doesn't have a math library, floating point numbers, etc.
- You're only limited to 16 local variables in a single function (including function parameters).
- The gas limit is around 8 million (as of April 17th, 2018).
- Non-linear activation functions are expensive to call.
- Neural networks are very computationally expensive.
- Storing things on the ethereum blockchain is expensive.
- And we noticed that there wasn't a protocol for exchanging ML models, even though there was some hype around ML + blockchain.
We're engineers, and we love solving problems. Running a neural network on the ethereum blockchain was a problem worth tackling.
Whats the objective of implementing neural network on block chain