Mr. Castillo's Machine Learning Questions

Question 1

Discuss the differences of the problem structure of face recognition and the game of chess.

Answer 1

main differences: additivity, difference in uniformity, and ability to make mistakes

Question 2

Discuss some similarities and differences between face super recognizers and chess grand masters.

Answer 2

Super recognizers are people who take up 1-2% of the general population with extreme ability to recognize faces. Chess grandmasters, on the other hand, are chess players with the highest title a chess player can obtain. As of today, there are 2,000 chess grandmasters living. For chess, it’s both nature vs. nurture, while super recognizers are more natural. Both recognize patterns.

Question 3

In the talk we described AI systems where there is an expert crafted evaluation function and where the evaluation function is learned from data. Discuss the strengths and weaknesses of each approach.

Answer 3

Handcrafted:
STRENGTHS: prove expert is brought in to formulate a linguistic strategy, relying on the presence of an expert; will make moves like humans
WEAKNESS: limited by the expert’s knowledge

Data approach:
STRENGTHS: picks up on hidden patterns (in chess, strange sacrifices that seem like playing against an alien); anyone who’s an expert in cog sci can write the algorithm (doesn’t require you to be a grandmaster)
WEAKNESSES: Relies on a huge amount of data; fail to plan (make moves based on the present state of the broad)