# The Geometry of Neural Network Landscapes: Symmetry-Induced Saddles & Global Minima Manifold

In this meeting together with Berfin Şimşek we discussed her paper with François Ged, Arthur Jacot, Francesco Spadaro, Clément Hongler, Wulfram Gerstner, and Johanni Brea “Geometry of the Loss Landscape in Overparameterized Neural Networks: Symmetries and Invariances” published at ICML 2021.

Interesting points:

- How the coefficient of the amount of saddle points (a_k) affects the calculations in the ratio?
- Can we make proper conclusions about easiness of conversion from the counts of the spaces or the overall space and how they relate to it should be taken into account?
- Do the calculations/results depend on the data distribution?

You can find the presentation that was held here.