Collaborative distributed machine learning

A computer-implemented method for distributed machine learning comprising:

• generating seed parameters of a machine learning model by performing preliminary training of the machine learning model at a model requester node that is an edge node of a network of cloud computing nodes;
• obtaining external updated parameters by facilitating one-step updates to the seed parameters by each of a plurality of other edge nodes;
• obtaining internal updated parameters by performing a one-step update of the seed parameters at the model requester node;
• aggregating the external and internal updated parameters at the model requester node;
• estimating learning utility of each of the plurality of other edge nodes, based on comparison of the external updated parameters to the internal updated parameters;
• requesting cost estimates from each of the plurality of other edge nodes;
• identifying a lowest-value edge node from the plurality of other edge nodes, based on a smallest value of a ratio of learning utility to cost estimate for each of the plurality of other edge nodes;
• reducing the plurality of other edge nodes by excluding the lowest-value edge node from the plurality of other edge nodes;
• generating a set of participating edge nodes from the plurality of other edge nodes by repeating steps of identifying and reducing until a total of the cost estimates from the plurality of other edge nodes is within a cost budget of the model requester node;
• and
• training the machine learning model by repeatedly:
• distributing most recent parameters of the machine learning model to the participating edge nodes;
• receiving updates to the most recent parameters from the participating edge nodes;
• and
• establishing new parameters for the machine learning model by aggregating the updates from the participating edge nodes.

A non-transitory computer readable medium embodying computer executable instructions that when executed by a computer cause the computer to facilitate a method of:

• generating a specification of a machine learning model at a model requester node that is an edge node of a network of cloud computing nodes;
• distributing the specification from the model requester node to a plurality of other edge nodes in the network;
• receiving replies to the specification from the plurality of other edge nodes at the model requester node;
• in response to the replies, the model requester node identifying a set of participating edge nodes that indicate ability to match the specification;
• and
• training the machine learning model, without exchanging training data among the model requester node and the participating edge nodes, by repeatedly:
• distributing most recent parameters of the machine learning model to the participating edge nodes;
• receiving updates to the most recent parameters from the participating edge nodes;
• and
• establishing new parameters for the machine learning model by aggregating the updates from the participating edge nodes.