I have an application that sends some data to a backend on a webserver where that data is being analysed and put through a neural network to come back with some other data that is being displayed on the client front-end.
Now, when a lot of clients are connecting simultaneously, we have to scale up, and then down again + there is network latency and you always need an Internet connection.
Because the backend isn't doing that much, besides just analysing and putting data through a neural network... it would be ideal to put this backend functionality in each client... no more need to scale up (every client has his own backend on board) + no more network latency.
The problem is of course that the most valuable thing is the trained neural network data, that in this case should also be present on each client... and of course I don't want it to be stolen as it costed a lot of money to buy training data for our neural network + then we had to train it as well...
So my question is:
Is there any way to run the backend neural network on the client side in a secure way, preventing our valuable trained neural network from being stolen?
Thank you for any advice!