From the course: TensorFlow 2.0: Working with Images
Unlock the full course today
Join today to access over 23,200 courses taught by industry experts.
Challenge: Experiment with hyperparameters
From the course: TensorFlow 2.0: Working with Images
Challenge: Experiment with hyperparameters
- [Instructor] We haven't had good results with the CIFAR-10 dataset using a neural network, so we're going to try two things. We're going to add two more layers to the neural network. At the moment, the neural network accepts the input image, which is of size 32 by 32 by 3. So if we flatten that, that's an input size of 3072 pixels in total, and then this goes on to a next layer, which has 128 nodes. Now, perhaps that's too big a jump and it isn't able to capture some of the changes in the image so effectively. So what if we gradually reduce that from 3072 pixels to 1024 and then 512 nodes? This might help us get better accuracy. In the second part of the challenge, we found that neural networks were great on the fashion eminence dataset, which is a dataset of grayscale images. Now the CIFAR-10 dataset has three channels, red, green, and blue. What if we just extract the red channel as input to the network instead of…
Practice while you learn with exercise files
Download the files the instructor uses to teach the course. Follow along and learn by watching, listening and practicing.