Conflict-Gated Gradient Scaling (CGGS)

How do you recover a full epidemic curve from just 20 noisy data points? In this video, we break down Conflict-Gated Gradient Scaling (CGGS) — a new method for training physics-informed neural networks (PINNs) on disease modeling problems.

We walk through:
- The SEIR model and how diseases spread through populations
- Why real-world epidemic data is sparse and noisy
- How PINNs combine data and differential equations — and why their gradients sometimes fight each other
- The Pareto deadlock problem with standard magnitude balancing (LRA)
- Our solution: a sigmoid-based gate that detects gradient conflict using cosine similarity
- A mathematical proof that the gate is self-correcting (bounded damage)
- Final results showing CGGS recovering the true infection curve

This work combines artificial intelligence, deep learning, and mathematical biology to improve how we model infectious diseases using limited public health data.

📄 Paper: Golooba & Woldegerima (2026) — https://arxiv.org/abs/2603.23799
🔬 Lab: DIMMS Lab — Disease-Informed Modelling, Methods & Systems
🏫 York University, Toronto, Canada

#PhysicsInformedNeuralNetworks #PINNS #AI #DeepLearning #ArtificialIntelligence #Epidemiology #DiseaseModeling #MathematicalBiology #GradientConflict #ComputationalBiology #PublicHealth #DataScience #NeuralNetworks #YorkUniversity #DIMMSLab #researchcentre

Видео Conflict-Gated Gradient Scaling (CGGS) канала DIMMS Lab