diff --git a/Workshop3/outline.md b/Workshop3/outline.md
new file mode 100644
index 0000000000000000000000000000000000000000..563e893938a94e047b84a9466077b2b64d06ff4b
--- /dev/null
+++ b/Workshop3/outline.md
@@ -0,0 +1,32 @@
+# Material graph neural networks for materials properties
+
+- Atoms molecules, crystals to graph format
+- State attributes, atoms, bonds
+- Use info to set up the model
+
+# Training NN to get desired material properties
+- Caloric materials:
+- Mechanical properties
+- Vibrational properties
+
+# Training for elastic constant
+- Bulk modulus
+- Start with the dataset (what dataset) to graph network
+
+# Set up training
+- Training, test, validation
+
+# Interested in? Perovskites, plastic crystals, etc. 
+
+- How better is the prediction?
+- Compare to the actual values
+
+- change from full materials project to a large subset to a small subset
+- see errors on predictions
+
+# Tuned by changing the dimensionality of the system
+- Efficiency vs accuracy
+- Cpus vs gpus 
+
+# Optimisation methods to predict properties where there is less data
+- How can we predict?