Hello. I read that, in addition to the implementation of neural networks in different systems, you will train them. What methods for training neural networks will you use?
There are several ways to train neural networks in Skychain
1. Closed-loop system. The developer independently develops a neural network and writes a script for its training, using examples of data from private datasets. Next, the developer uploads the neural network into a closed loop. The neural network's code should implement several typical methods, such as training, inference so that this neural network could be used in a closed-loop system. Next, the closed-loop starts the training process of the neural network using the full volume of data and keeps the neural network in protected storage. The neural networks remain available to consumers through the Inference method.
2. The developer himself trains the neural network on his computer in any way and simply connects it to Skychain for the purposes of inference.
3. The developer can train a neural network on his own dataset, using the distributed training system, provided by Skychain. In this case, the developer must implement several additional methods that will combine the neural network parameters after each iteration of the distributed learning process.
To sum it all up - the developer himself chooses the libraries, algorithms, etc. used and should simply implement a number of public methods in order for his neural network to be uploaded and used in the Skychain system.
Show Posts
