Mystery Machine Learning: Classifying Text with RNNs, Keras, and Scoob and the Gang

on YouTube ↗46m

" summary

Builds a recurrent neural network with Keras/TensorFlow on a CSV of Scooby‑Doo lines tokenized and embedded, then serves the model on RedisAI and queries it from a JavaScript REST endpoint. The demo classifies lines like "jinkies" → Velma and "look out scoob" → Shaggy, shows training metrics rising but test accuracy collapsing (overfit) and highlights the vanishing gradient issue; model works for catchphrases but fails generalization.

AI-written summary · about

" chapters (14)

1. 0:00Introduction and Scooby-Doo hook
2. 1:31Talk overview and disclaimers
3. 3:56Scooby-Doo and data science analogy
4. 5:41Problem setup and dataset
5. 6:45Neural networks explained
6. 11:18Single neuron deep dive
7. 13:26Activation functions
8. 19:02Recurrent neural networks
9. 25:18Text classification with RNNs
10. 26:22Tokenization and embeddings
11. 33:20Keras code walkthrough
12. 37:34Demo and model training
13. 39:16Redis AI deployment
14. 42:44Summary and resources

" related talks

• 36mLearning about machine learning from kidsDale Lane · NDC Porto 2022
• 52mBreaking black-box AIEvelina Gabasova · NDC London 2020
• 42mThe Machine Understands Text — Learning and Search at Micro and Macro ScaleŁ. Kuczyński · 4Developers 2021
• 39mLive Coding A Machine Learning Model from ScratchSara Robinson · Google I/O 2019
• 52mMachine Learning Crash CourseSamuel Taylor · Windcity DevFest 2019
• 01h 30mIntroduction to Deep Learning & Natural Language ProcessingRaghotham Sripadraj, Nischal HP · PyData London 2016
• 54mMachine Learning for DevelopersDanilo Poccia · Voxxed Days Zurich 2017
• 29mMachine learning for front-end developersCharlie Gerard · JSHeroes 2019