Federated learning with scale-mamba

[Idoia Gamiz]

Hi! I have some doubts about theoretical terms. I use scale-mamba in order to train machine learning algorithms in this way:

1- From each player "i" I collect the data that will be used to train the algorithm: X_train_i, y_train_i.

2- I join all the X_train_i and all the y_train_i in the same matrices: X_train, y_train.

3- I use these joint matrices to train my algorithm and update the weights by gradient descent.

By its definition Federated Learning usually consists in giving the model to each party and each party calculating its own weights, and then putting all this weights together. It is not exactly my case.
Some also consider that when you apply multiparty to train a machine learning algorithm it is actually federated learning, (which is what I consider I am doing). Is it correct?

Thank you in advance.

[Nigel Smart]

Federated learning "usually" means a process in which different parties
train a model on their own data, they then "pool" their models to obtain
a global model [in some sense]. Note that the parties perform DISTINCT
local computation. The pooling "can" be done by MPC, but that is
not required

If you compute a single joint model on data, using secret shared data
[which is kept secret] then that is an MPC computation which does
standard machine learning.

What you seem to be doing is the latter, and not federated learning

Nigel

--
You received this message because you are subscribed to the Google Groups "SPDZ/SCALE-MAMBA Discussion Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spdz+uns...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/spdz/7720825d-d809-4312-8aac-a34713f0e70en%40googlegroups.com.

-- 
Nigel Smart