AutoML Explanation

Dear DLS team,

What kind of AutoML are you providing?
Can it be controlled by user?
Can user get insights about why AutoML behaves so with the user’s model?

We have implemented a custom algorithm which design the network based on the best practices for given type of dataset (input/output type).

The generated output (layers in the model) is fully controllable as if you have designed that model.

Can you clarify what you meant by “why AutoML behaves” ? Are you referring to the model generation behavior or the generated model behavior ?

If it is latter, then you can look at all the layers and its configuration (you can look at code view to get a summary). The generated model behavior only depends on the layers selected and its parameters.

Hi, @rajendra

I am referring to the model generation behavior - why generator decides to use concrete combination of layers and connections between them. If you are using some RL approach working with representation of NN(to be generated) encoded as the some sort of genome, this will be extremely interesting topic to discuss 8]

Hi @Ambreaux ,

At this point, we prefer not to discuss AutoML model generation behavior. Sorry about that.

Ok, no problem if you are under NDA =]