Figure 6
Schematic of deep transfer learning approach. \({\mathcal {D}}_S\) refers to input data from a source domain, in this case a HAR dataset, to learn a task \({\mathcal {T}}_S\), which is represented by the label space \(\mathcal {Y}_S\) (the HAR activity classes). \({\mathcal {D}}_T\) refers to the target domain, in this case the FLOODLIGHT data, where \(\mathcal {Y}_T\) are the disease classification outputs of HC, PwMSmild or PwMSmod for target task \({\mathcal {T}}_T\). During transfer learning, a model’s parameters and learned weights, \(f(\cdot )\) of \({\mathcal {D}}_S\), are then used to initialise and train a model on target domain \({\mathcal {D}}_T\) and task \({\mathcal {T}}_T\). Transfer learning is then performed by transferring the source model’s layers (where these weights and parameters are “frozen”) to subsequently re-train a new model (i.e. fine-tune) using \({\mathcal {D}}_T\) data for the new target task, \({\mathcal {T}}_T\). Downstream layers in the network are fine-tuned towards this new target task decision \(\mathcal {Y}_S\).
