Hey Outliers,
Last town hall David went over his research on self-organizing maps, specifically Kohonen Self Organizing Maps (KSOM). KSOM is a method of representing higher dimensional data in much lower dimensional spaces. It is almost like a combination of Principal Component Analysis and K-Means. KSOM can be used to reduce the dimensions just like Principal Component Analysis but in a non-linear way. Just like K-Means, KSOM clusters data by grouping together nearby points. However KSOM does NOT need to know how many clusters there are. It learns the ideal number of clusters (given the right parameters). Best of all (being biased) it is a Neural Network! During the presentation, David successfully demonstrated representing higher dimensions into 2D spaces. He also visualized and explained clearly what happens to decision boundaries when certain hyperparameters are not tuned correctly.
Learn more about Kohonen Self-Organizing Maps
Community Updates
Book Study
Mathematics for Machine Learning finished!
Probabilistic Machine Learning
Starting the book on Monday, September 26 in Discord!
Interpretable Machine Learning
Starting Chapter on Decision Trees
Applying what we learned on NBA data
Hands on Machine Learning
Tweet of the Month
Statistics Links
Video of the Month
Jobs
If you are a lead/head-level person looking for a data science role, feel free to hit up leo_lmre in Discord!
Photo by Marcus Urbenz on Unsplash
Thank you for participating in this town hall. The next one is in Oct 27, 2022 at 8pm EST.