Is it a predictive model? Is it causal inference? Well... It is running a greenhouse.
summary
Automation of greenhouse irrigation at Source combines concrete claims and tools: a multi stage approach first trains a predictive model of water uptake on non irrigation periods then subtracts those predictions to estimate the causal effect of irrigation shots, using LightGBM and linear regression and experimenting with double machine learning via econml The ground truth from scales where one liter equals one kilogram lets evaluation show the orange multi stage estimate approaches reality after correcting for leakage unlike a naive black box which sometimes gets lucky predictive then causal
AI-written summary · about
chapters (23)
- 0:00Causal vs predictive thinking
- 1:09Predictive-to-causal trap
- 2:55Greenhouses and food security
- 5:03Irrigation in greenhouses
- 7:32Irrigation data saturation
- 8:54Causal vs predictive in irrigation
- 9:59Testing black-box models
- 11:23Double machine learning
- 13:30Feature selection for causality
- 14:32Lag features dangers
- 15:37Treatment-induced leakage
- 16:19Identifying causal components
- 18:25Safe predictive components
- 19:07Beyond black-box modeling
- 20:34Key takeaways
- 21:17Q&A control groups
- 23:02Q&A Kalman filters
- 24:48Q&A stopping time
- 25:54Q&A plant variability
- 26:58Q&A A/B testing feasibility
- 28:45Q&A observational data
- 30:52Q&A physics vs ML
- 32:16Q&A systems and control
related talks
25mFrom seed to store: AI to optimize the indoor farms of the future
19mPlant Factory: Sensor, Data, Machine Learning
31mAI-supported transformation of the urban infrastructure
49mModeling our future: Advancing climate research with AI
31mPython and the Holy Grail of Causal Inference
43mBeyond Just Prediction: Causal Thinking in Machine Learning
31mBridging IoT and Machine Learning with Python for Sustainable Agriculture
09mGen AI Workflows for Climate Risk Modeling in Food Systems