Introduction to Deep Learning · HIT

Week 10   Part III · Architectures & Representation Learning

Recurrent Networks (RNNs)

Sequence data and recurrence; the RNN cell; backpropagation through time; vanishing and exploding gradients.

Curated, free, canonical references for this week: a course or lecture, a book chapter, a video, and an authoritative blog post or official tutorial. Each opens in a new tab.

Course
Stanford CS224n: Natural Language Processing with Deep Learning web.stanford.edu

Its Language Models and RNNs lecture covers recurrence and the vanishing-gradient problem.

Book
Dive into Deep Learning, 9.7 Backpropagation Through Time d2l.ai

Derives BPTT by unrolling the RNN graph and shows how vanishing/exploding gradients arise.

Video
MIT 6.S191: Recurrent Neural Networks youtube.com

Accessible lecture introducing sequence data, the RNN cell, recurrence, and BPTT.

Blog / Docs
Andrej Karpathy: The Unreasonable Effectiveness of Recurrent Neural Networks karpathy.github.io

Classic post building intuition for what RNNs learn via character-level language models.

← Back to the Week 10 lab

PreviousWeek 9: Convolutional Networks IINextWeek 11: LSTMs, GRUs & Sequence Tasks