2/16/18

Watched a video on an issue with recurrent neural networks called the "Vanishing Gradient" problem. I won't get into the mathematics of it, but the basic idea of it is that the technique used to update a recurrent neural network's weights is extremely ineffective with traditional methods because the network's accuracy isn't able to improve much after each training example. Next, I watched a lecture on LSTMs (Long Short-Term Memories), which were extremely complicated. Even after the lecture, I wasn't convinced that they would be useful for my project, so I watched a few more videos on them on YouTube. I wasn't convinced; it still seems like a simpler artificial neural network is the more practical way to go. After watching this deep learning content I decided to try to learn how to create a clickable button in python.

Comments

Popular Posts