Abstract
We propose a topological description of neural network expressive power.
We adopt the topology of the space of decision boundaries realized by a neural architecture as a measure of its intrinsic expressive power.
By sampling a large number of neural architectures with different sizes and design, we show how such measure of expressive power depends on the properties of the architectures, like depth, width and other related quantities.
We adopt the topology of the space of decision boundaries realized by a neural architecture as a measure of its intrinsic expressive power.
By sampling a large number of neural architectures with different sizes and design, we show how such measure of expressive power depends on the properties of the architectures, like depth, width and other related quantities.
Original language | English |
---|---|
Number of pages | 1 |
Publication status | Published - 11 Dec 2020 |
Event | Topological Data Analysis and Beyond. : Workshop at NeurIPS 2020 - Virtual Duration: 11 Dec 2020 → 11 Dec 2020 Conference number: 2020 https://tda-in-ml.github.io/ |
Conference
Conference | Topological Data Analysis and Beyond. |
---|---|
Abbreviated title | NeurIPS |
Period | 11/12/20 → 11/12/20 |
Internet address |
Keywords
- Neural Networks
- Expressive Power
- Decision Boundary
- Classification