Playing Newton: Automatic Construction of Phenomenological, Data-Driven Theories and Models

on YouTube ↗51m

" summary

Automated inference builds phenomenological, data-driven theories as nested ODE families (sigmoidal-network and S-system bases) and scores them with Bayesian model selection; it finds that two dynamical variables are necessary for planetary motion, reduces a 7-species glycolysis model to 6 effective variables (carbon conservation), and produces a two-variable model of C. elegans escape that predicts the recoil-to-forward transition time versus laser intensity. The method fits hidden-variable dynamics and generalizes to held-out conditions.

AI-written summary · about

" chapters (4)

1. 0:00Start
2. 0:47What do we study? . How biology processes information and learns?
3. 5:30At CNS 2014 • A common opinion: The final theory of biological systems will be a large multiscale computational model. We need more experimental data to specify its detais.
4. 15:06Armed with this definition of a good theory: • Do there exist good theories / models of complex biological systems?

" related talks

• 01h 11mHidden Physics Models: Machine Learning of Nonlinear Partial Differential EquationsMaziar Raissi · AAAI-MLPS 2020
• 01h 03mGeneralization from the behavioral perspectiveMax Raginsky · Emerging Generalization Settings 2024
• 33mTwo Defeasible Challenges for algorithmic Scientific DiscoveryBenjamin Jantzen · Thoughtworks E4R Symposium 2020
• 32mData-Driven Inverse Modeling with Incomplete ObservationsKailai Xu · AAAI-MLPS 2020
• 34mData-Driven Solutions in EngineeringVladislav Rosov · Data Science fwdays'19
• 57mDarwinian Machine Learning: Principles of Machine Learning in Evo-devo, Evo-eco and Evolutionary...Richard Watson · Computational Theories of Evolution
• 49mPedro Domingos, Professor, University of WashingtonPedro Domingos · MLconf ATL 2015
• 01h 10mBayesian Optimization and Machine Learning for Accelerating Experiments in the Physical SciencesStefano Ermon · AAAI-MLPS 2020