R16 Questions Flashcards
According to this author, what was the most exciting feature of the Netflix competition to recommend movies? What does this tell you about the biggest challenge to machine learning approaches that learn by “experience”?
It offers large amounts of real data.
The biggest challenge is that the machine learning approaches works better with more data to learn.
Finding attributes that are good predictors of behavior is the goal of the training phase. What is somewhat odd or disturbing about the attributes?
Attributes may be hard to interpret (quite subtle and incomprehensible).
What is reinforcement learning? Give an example of its use.
Reinforcement learning is trial and error learning.
One example of this is when playing a game like tic-tac-toe.
What is non-supervised learning? Give an example of its use.
Unsupervised learning groups based on similarity. There are no labels (e.g. “right answers” to the data), which makes it much harder to perform.
example = clustering
What is over-fitting and how do researchers guard against it?
Overfitting: when your model overanalyzes your data and identifies patterns that are not relevant.
To avoid this, algorithms are biased to keep the models as simple as possible with regularization or tested on new data by validating.
