New answers tagged

1

The advantage of using a pre-trained model without loading the weights (which would mean you are only use the model, not a pre-trained version) is that you can easily use an existing model architecture and applying it to your problem. This can save you quite some time since you don't have to build the model architecture yourself in tensorflow/keras/pytorch ...


1

You can first write the bottleneck features into a tfrecords file, and then load them as a dataset for the training phase. In the tensorflow documentation you can find complete examples of how to do both.


0

Yes, it can depend on it because it changes the data distribution with the network is trained. You shouldn't consider random seed as a hyperparameter. Keep the same random seed and run comparison. Do this for at least 5 or 10 random seeds. You will definitely have a winner. If you don't go for more than 10 unless you get a winner. That will give you ...


1

In the context of that paper, pre-train then fine-tune on the same dataset does not really make sense, as the pre-training is unsupervised, and the fine-tuning is with labelled data. But, generally, if you have trained on dataset X for N epochs, and then you fine-tune one more epoch using the whole of X, it is just another way of saying you trained for N+1 ...


0

Yes, the typical approach is to obtain the saliency map of the input, which are "heatmaps" of the contribution of each pixel to the final classification. In this free online book about Explainable ML, you can find the most relevant approaches to obtain saliency maps, like vanilla gradients, together with other pixel attribution techniques. Here you ...


Top 50 recent answers are included