A Learning Story: Love, Scala & Artificial Intelligence

on YouTube ↗39m

" summary

Implements a denoising autoencoder that encodes Scala Meta ASTs into a latent vector using Tree LSTMs and decodes with stacked LSTMs, trained in a PyTorch pipeline. Pipeline: Scala source → Scala Meta AST → JSON token lists → token embeddings → tree-to-vector encoding → sequence decoding; data is sharded into many JSON files for scale. Test-case reconstructions show low cross-entropy loss and some generated code compiles; production-scale training and Scala-native bindings remain future work.

AI-written summary · about

" chapters (16)

1. 0:00Learning journey intro
2. 1:17Disclaimers & context
3. 2:21Legacy code struggles
4. 4:40Scala macros & contribution
5. 5:49Neural networks understand code
6. 7:35Autoencoders for code
7. 13:03Scala source code analysis
8. 15:58Vanilla RNNs limitations
9. 18:13LSTMs solve vanishing gradients
10. 22:13Tree LSTMs for code
11. 24:47Data prep for neural nets
12. 29:23Encoding source code
13. 32:40Decoding sparse code
14. 33:23Autoencoder architecture
15. 34:54Results & future work
16. 36:41Acknowledgments

" related talks

• 54mWhy Netflix ❤'s Scala for Machine LearningJeremy Smith, Aish · Scala Days Lausanne 2019
• 39mMachine Learning with Scala on SparkJose Quesada · Scala Days Berlin 2016
• 40mScala and Spark in the New Deep Learning EraEmiliano Martinez Sanchez · Scala Days 2023 Madrid
• 39mDive into Deep Learning with ScalaSören Brunk · Scala Days Berlin 2018
• 23mRethinking scalable machine learning with Spark ecosystemAdi Polak · Scale By the Bay 2021
• 35mJVM based DeepLearing on IoT data with Apache SparkRomeo Kienzler · Jazoon Zurich Autumn 2017
• 45mDeep Learning and NLP with SparkMelanie Warrick, Andy Petrella · Scala Days Berlin 2016
• 32mMachine Learning Infrastructure at Stripe: Bridging from Python JVMRob Story · PyData Seattle 2017