Tutor profile: Jazz S.
Subject: Machine Learning
What does it mean for a model to overfit?
If a model has overfit, that essentially means it's figured out a way to hack the system -- the model has unfortunately learned to utilize irrelevant information specific to the training (or in some cases, validation set) which helps it attain a really high accuracy in training, but causes poor predictions on the test set. A strong practical signal of overfitting is an increase in loss on the test set while the loss on the training set is still decreasing. In terms of theory, overfitting relates to the bias-variance tradeoff, where models that have overfit have high variance and low bias (which is a really cool topic we can discuss more together later!). There are lots of methods to prevent overfitting, varying depending on the machine learning approach you're using. One common strategy to deal with overfitting in neural networks is to simply reduce the network size, thereby preventing the model from learning something needlessly complex (and thereby likely noisy, if the model is overfitting).
Subject: Computer Science (General)
What's the complexity of a selection sort, and how does it work?
Great question! A selection sort is one way to sort a sequence of numbers. Basically, you go through the sequence of numbers, find the smallest number, swap that with the first element of the sequence, go through the remaining sequence, find the smallest number, swap that with the second element of the sequence, and repeat until you've swapped the second to last element of the sequence. The fundamental idea is that you're going through and placing the first smallest, second smallest, third smallest, etc. elements where they should be in the sequence -- in the first, second, third, and remaining positions. If there are N elements in the sequence, notice that in order to swap an element to the first slot in the sequence, you have to read through the entire sequence, which takes linear O(N) time (this is Big-O notation!). For the second element, you have to read through the remaining N-1 elements, which also takes linear O(N) time. If we follow the pattern here, we notice that there are N different slots, each requiring O(N) time to complete. This means the runtime of a selection sort is N * O(N) = O(N^2), so a selection sort takes quadratic time. This isn't the most efficient way to sort a sequence of numbers -- can you think of a faster algorithm? (Hint: Experiment with recursion)
Subject: College Admissions
How can I maximize my chances of getting into a good university?
There are five pillars that are most important in your applications: your GPA and transcripts; your standardized testing scores*; your essays and personal statement; your letters of recommendation; and any extracurricular activities or passions that demonstrate the compelling perspective you bring to the table. Doing exceptionally well in all five dimensions of course increases your chances almost to certainty to being accepted into your dream college, but if you feel weak in one or the other then there's no need to worry! Other pillars, as well as the knowledge of how to craft and frame your application overall, can result in a winning profile. *Note: COVID has affected this factor :)
needs and Jazz will reply soon.