CS224n

#concept-pamphlet #todo: take from the class what i want

Cheatsheet §

Create flashcards for every term below. Be concise and use equations where possible. Keep your answer succinct, accurate, and use math equations where possible.

• word vectors
• back propagation
• neural networks
  • encoder
  • decoder
• dependency parsing
• neural networks
  • recurrent neural networks
    • fancy version
  • recursive neural networks
• language models
  • n-gram
• vanishing gradients
• seq2seq
• machine translation
• attention
• subword models
• transformers
  • BERT
  • ELMo
  • transfer learning
• pre-training
  • knowledge-pretrained language models
• question answering
• natural language generation
• bias in models
• retrieval augmented models + knowledge
  • LaMDA
• convolutional neural networks
• tre recrusive neural networks
• constiutency parsing
• scaling laws for language models
• coreference
• editing neural networks

Reference §

1. https://web.stanford.edu/class/cs224n/index.html#coursework

Skeleton §

levenshtein distance equation ? it is the minimum number of single-character edits (insertion, deletion) required to change one word into another. tells you how different 2 strings are. the higher the number, the more different a string.made in 1965 source: https://en.wikipedia.org/wiki/Levenshtein_distance

---

other helpful stuff

• practical: using hugging face transformers

---

https://twitter.com/yimatweets/status/1578082015668146177?s=46&t=iKJdkAuqqJUYFrftwrlOHA

• “Can principles lead to purely explainable whitebox deep networks that work the best in all settings: supervised, incremental, unsupervised, classification, generative, and autoencoding, with smaller networks and less data/computation?”

https://web.stanford.edu/class/cs224n/index.html#schedule

Concepts

• Bag of words
• K nearest neighbor 

related, ML engineering

• Linear regression
• Logistic regression
• Evaluation metrics
• Docker
• web services
• cloud
• Model deployment
• Tree-based models
• Neural networks
• Kubernetes