By using this website, you agree to our privacy policy [ Ok ]

Course review: NLP with probabilistic models

by Abdur-Rahmaan Janhangeer

Stats is always great! I liked this course, no wonder it has a 4.7/5 rating, the highest of all the 4 in the specialization.

week 1: Essentially Norvig’s explanation on how Google built it’s spell corrector. Norvig’s explanation is better in my opinion. This is one great illustration that problem solving beats knowledge hoarding where Norvig was amazed his maths peers had no hint on how it worked.

week2: Markov models, emission matrices and lots of viterbi. Know and code.

week3: Ngrams and dealing with unwanted cases. Know and code.

week4: Very exciting. The part about creating embeddings from scratch is 🔥 . However, they did a very very very poor job at explaining classical concepts like gradient descent. it’s not because that they did not code it from scratch, it’s because if they are really explaining it for the first time, it’s bad at making people understand what it is. Relu and softmax were top notch though. I think i followed most vids with attention, very engaging!