Day 4: Stanford Lecture and the softmax Function

Today I watched a Stanford lecture on loss in machine learning and cost optimization function. Loss functions are a crucial part of neural networks. I learned the basics of feed forward neural networks from a helpful web series: (https://www.youtube.com/watch?v=ZzWaow1Rvho&list=PLxt59R_fWVzT9bDxA76AHm3ig0Gg9S3So) . After watching the YouTube series I switched to the Stanford lectures to learn more on loss functions. The purpose of a loss function is to update your neural networks classifying function. By giving your predicted result and comparing it to the actual result you are able to tell where your model went wrong and you can then update it to improve over the next iteration. I learned about hinge loss and the squared loss function which are two different ways to calculate your error. Using this knowledge I was able to begin to explore the soft max function. I don't fully understand the math behind it today, and I will continue to look into it tomorrow. The image below is a visualization of a neural net that used soft max . Today I learned an incredible amount about feed forward neural networks how they utilize loss functions in order to update their predictions.


Source: http://cs231n.github.io/neural-networks-case-study/

Comments

Post a Comment

Popular posts from this blog

Day 11: Network success and an interesting lecture

Image
I started today debugging my code for my neural network. After a healthy amount of googling and some perseverance I was able to roughly train and evaluate my model. The performance was subpar but hopefully I will be able to improve that next week. When that was over I went to the free pizza lunch provided for all of the REU summer research students and I then attended a lecture on visual perception. Although this was not very related to my project I still found it to be a valuable use of my time. It was extremely mind-blowing to realize that our eyes can only see details in a very small field of vision. Our peripheral vision is a lot worse than I previously thought. One of the coolest visual demos was these two tables shown below. Believe it or not the tabletops are exactly the same size and shape. Human visual perception is fascinating and I plan to continue attending the lectures throughout the summer. Source:  http://www.optical-illusionist.com/illusions/table-size-illusio
Read more

Day 26: Polishing my presentation

While I was waiting for my experiments to run on the server I spent the day presenting in front of various members of kLab. I got some very good advice from Dr. Kanan in the morning which I used to update my presentation. Later in the day with the help of  Tyler, Dr. Kemker and Kushal I was able to get a presentation that I was happy with. I plan to update the presentation with the results from the experiments I ran.
Read more

Day 22: Implementing new lifelong learning models

Today I implemented a reservoir sampling algorithm as well as a queue nearest neighbor program. The reservoir algorithm functions similarly to the incremental nearest neighbor program i implemented on Friday. The only difference is that it only holds on to a certain amount of data points per class and when new data points are added it removes an existing point at random. This is an effective way to update your prediction function and it uses less memory than the standard nearest neighbor function. The other program I implemented was the queue nearest neighbor program. This is similar to the reservoir program. The thing queue does differently is that it removes the oldest data point for a specific class when new data points are added. This is also an effective way to reduce memory usage. Tomorrow I plan to combine the VQA data set and these classifiers in order to test the effectiveness of lifelong learning on VQA.
Read more