The Modern Mathematics of Deep Learning explained for non-mathematicians
Understand the mathematics of deep learning through small, hands-on examples, following the original research paper.
Start readingHow does a model identify a penguin?
Training here means measuring each species’ ranges and average. The model then picks the species whose average is closest.
This lab classifies by nearest species average. The Concepts project trains weights with softmax. Same penguin measurements; different classification methods. Predictions can differ.
Real measurements; a nearest-centre classifier you could compute by hand, not a deep network. Data: Kristen Gorman / Palmer LTER; palmerpenguins, CC0
- ● Adelie
- ▲ Chinstrap
- ■ Gentoo
Pick a penguin
- Bill length (mm)
- 40.6
- Flipper length (mm)
- 193
Penguin 1 of 3
| Species | Penguins | Bill range (mm) | Average bill | Flipper range (mm) | Average flipper |
|---|
Train the model to compute ranges and averages from the 240 training penguins.
An independent educational explanation of The Modern Mathematics of Deep Learning — Julius Berner, Philipp Grohs, Gitta Kutyniok, Philipp Petersen. Examples use invented numbers unless a measured dataset is credited.
Chapters in paper order
Read from the beginning, or choose a chapter.
How learning works
How can numbers become a prediction?
Beyond the practice set
How can a large network learn without memorizing everything?
Why add more layers?
What difference does another layer make?
So many inputs!
How can a network handle many input dimensions?
Finding better settings
How does it find useful parameters?
Designing a network
What changes when connections change?
Finding useful patterns
Which patterns does the network recognize?
Learning meets science
How can a measurement reveal something hidden?