What is Transfer Learning? As human we know and practice related knowledge from past learning and using it to the new one. The more related our previous knowledge with the new job, the easier we can become proficient. The idea is same with Transfer Learning which knowledge learned in one or more source tasks is transferred and used to improve the learning in the target task.
Let dive in a little deeper on how Transfer Learning work for example in image identification. First we need to delete what called as “loss output”. Loss output is the last layer before making the predictions. Then we will replace the old loss output with the new one.
Next we would take the smaller data set for the image and improve it on the neural network or the last few layers or only the loss layer. By implementing this techniques the output on the new Convolutional Neural Networks (CNN) will be the image identification.
Beside image recognition, Transfer Learning also can be used as speech recognition. Recurrent neural networks usually used for this task. To do this we need two same speech-related data set and use the same techniques on CNN.
E-SPIN in the enterprise technology solution trading and maintenance business. Feel free to contact E-SPIN for your operation or project requirements, include implementing cutting edge and latest emerging technologies for breakthrough enterprise business performance and digital transformation, include adopting of transfer learning into Machine Learning context.
