It’s been a while since I posted. I’ve continued to mess around with Machine Learning. As a seemingly natural extension of Andrew Ng‘s class, I took Geoffrey Hinton‘s Neural Networks for Machine Learning class. I have to say, it totally blew my mind. If we haven’t already gotten to a near-sentient AI, I’m convinced that we’re very close.
For my own exploration of Machine Learning, I’ve decided to work through the family of neural network types in a similar order to the sequence presented in Geoffrey Hinton’s class. That means, that I start with the very first type of neural network, the Perceptron.
A Perceptron takes multiple input values and makes a single logistic decision about it. I took a collection of MNIST digits to see how well a Perceptron would do at identifying them. You’d imagine that this would be a natural fit for a Perceptron as digits are reasonably distinct… but I discovered that there really is not enough information stored within the weights of a perceptron to make it an effective classifier for MNIST.
Here’s what the perceptron weights looked like for number 3. White means positive coefficients and black means negative. Gray means near-zero. The images read from left to right and represent the state after increasing batches of 1000 labeled cases.
Here you can see that the more batches you run against the perceptron, the more complex the weights get. These complexities are probably overfitting. I thought, meh. Let me see how well it performed. Now, as 3s only represent about 10% of the label data, if I always guess “Not a 3″, I would see a 90% correct guess. This means that I wanted a value higher than 90%. It turns out that my classifier scored 90.26% which caused me to suspect that my perceptron had guessed all 0s. From actually looking at the guesses, though, I realized that no… the perceptron was actually guessing based on data, so I needed to dig a bit more into what it was doing. Here’s the distribution of its answers:
Correct Negative: 4034 Correct Positive: 479 False Negative: 466 False Positive: 21
So for real, it was guessing right about 90% of the time. What screws it over ends up being its false negatives. blah.
I decided to try something. Instead of weighting positive and negative cases equally when determining weights, I thought that I would add a term called alpha which weights positive and negative cases differently. In this case, since the positive case only happens one in 10 times, I wanted to weight the negative cases by 1/10th, so I set alpha to 0.1.
Here is an image of the coefficients using the values weighted by alpha. You can see that, especially in the earlier batches that the shape of the 3 is a lot fuzzier and we have less complexity, even in the later iterations. So how well did THIS set of coefficients work?
Correct Negative 3962 Correct Positive 472 False Negative 538 False Positive 28
It seems like it did a lot worse. It had a lot more false negatives. The false positives and correct positives also dropped but not by much. It seems like the fuzzier coefficients caused the perceptron to be less certain about a particular trial.