Skip to main content
. 2018 Nov 28;38(4):1800108. doi: 10.1002/minf.201800108

Figure 3.

Figure 3

An example of neural network model using “soft parameter” sharing. Two networks are trained in parallel for each individual task. The soft parameter sharing is done by introducing a penalty function, which prevents neural network weights in both models from differing greatly, as well as by regularising neural network features at the last layer. Reprinted from ref. [25] under the Creative Commons license CC‐BY 4.0.