Uploaded March 2020 | Updated November 2026, 3 hours ago
slides: mlvu.github.io/lectures/61.SequentialModels.annotated.pdf
course materials: mlvu.github.io
errata: In slide 20, one of the factors of the decomposed joint probability is missing.
In the last lectures, we will discuss various deviations from the standard offline machine learning recipe we've discussed so far. Today: learning sequences. Language, music, time series and so on.
In the first half, we discuss the simple, but powerful approach of the Markov Model, and our first embedding model: word2vec.
In the second half, we discuss recurrent neural networks. A very powerful, but a bit more complex approach to dealing with sequences. Specifically, we focus on the LSTM; one of the most popular neural networks in use today.
slides: mlvu.github.io/lectures/61.SequentialModels.annotated.pdf
course materials: mlvu.github.io
errata: In slide 20, one of the factors of the decomposed joint probability is missing.
In the last lectures, we will discuss various deviations from the standard offline machine learning recipe we've discussed so far. Today: learning sequences. Language, music, time series and so on.
In the first half, we discuss the simple, but powerful approach of the Markov Model, and our first embedding model: word2vec.
In the second half, we discuss recurrent neural networks. A very powerful, but a bit more complex approach to dealing with sequences. Specifically, we focus on the LSTM; one of the most popular neural networks in use today.










