Lawful AI

on YouTube ↗21m

" summary

Argues that Scala’s FP libraries rarely serve data scientists and that building lawful libraries using abstract algebra (Spire, HLearn patterns) yields concrete wins: fast nearest neighbor for arbitrary metric spaces, parallel batch and online training and cross-validation. Gives a real bug where an H2O model’s entropy consumption as a side effect produced week-to-week noise and would have been caught by clearer signatures. Proposes implementing Pyro-style probabilistic algorithms with Spire and notes Polytope uses Shapeless to make genetic programming type-safe.

AI-written summary · about

" chapters (7)

1. 0:00Introduction and background
2. 4:08Lawful libraries in Scala ecosystem
3. 7:02Reasons for lawful AI libraries
4. 12:23Cross-field communication challenges
5. 15:57Category theory across disciplines
6. 17:00Future of algebraic structures in AI
7. 19:48Q&A and closing remarks

" related talks

• 47mScala: The Unpredicted Lingua Franca for Data ScienceDean Wampler · GOTO Chicago 2016
• 47mScala: The Unpredicted Lingua Franca for Data ScienceDean Wampler, Andy Petrella · Scala Days Berlin 2016
• 29mResponsible AI Challenges and RecommendationsRicardo Baeza-Yates · Scale By the Bay 2021
• 39mMachine Learning with Scala on SparkJose Quesada · Scala Days Berlin 2016
• 24mPractical Probabilistic Programming with FigaroAvi Pfeffer · MLconf 2016 Seattle
• 52mDoubt Truth to be a Liar: Non Trivialityy of Type Safety for Machine LearningMatthew Tovbin · Scala Days Copenhagen 2017
• 32mStructured Deep Learning with Probabilistic Neural ProgJayant Krishnamurthy · Scale By The Bay 2018
• 01h 07mProbabilistic Programming and Readable ModelsFritz Obermeyer · PyData Yerevan 2022