Penny is a simple tool to help us understand what wealth and poverty look like to artificial intelligence algorithms. The tool lets you change the landscape of a city, by adding and removing urban features like buildings, parks and freeways to high-resolution satellite imagery. With this interface, you can explore what different kinds of features make a place look wealthy or poor to an AI.
9. Satellite Imagery & Census Data Neural Network
The Training Model (Resnet 50)
10. Satellite Imagery & Census Data Output
=
Neural Network
The Training Model (Resnet 50)
Q0 = 91%âš
Q1 = 5.64%âš
Q2 = 2.55%âš
Q3 = .41%
11. Satellite Imagery & Census Data Output
=
Neural Network
The Training Model (Resnet 50)
The model carriesâš
what it has learned and âš
repeats the process.
Q0 = 91%âš
Q1 = 5.64%âš
Q2 = 2.55%âš
Q3 = .41%
12. The Training Model (Resnet 50)
Input Normalization Normalization Normalization MergeActivationActivationConvolution Convolution Convolution
Neural Network
The data is fed into the model.
13. Input Normalization Normalization Normalization MergeActivationActivationConvolution Convolution Convolution
The Training Model (Resnet 50)
Neural Network
A number of ïŹlters are applied to the image.
14. Input Normalization Normalization Normalization MergeActivationActivationConvolution Convolution Convolution
The Training Model (Resnet 50)
Neural Network
The resulting new values are normalized to be within
learned mean and standard deviations of the dataset.
15. Input Normalization Normalization Normalization MergeActivationActivationConvolution Convolution Convolution
The Training Model (Resnet 50)
Neural Network
Separates out features from important and non important ones.
16. Input Normalization Normalization Normalization MergeActivationActivationConvolution Convolution Convolution
The Training Model (Resnet 50)
Neural Network
Repeat.
17. Input Normalization Normalization Normalization MergeActivationActivationConvolution Convolution Convolution
The Training Model (Resnet 50)
Neural Network
Results are merged with previous iteration.
32. âDespite this encouraging process, there is still little
insight into the internal operation and behavior of
these complex models, or how they achieve such
good performance. From a scientiïŹc standpoint, this is
deeply unsatisfactory. Without clear understanding of
how and why they work, the development of better
models is reduced to trial-and-error.â
Visualizing and Understanding Convolutional Networks - Matthew D. Zeiler, Dept. of Computer Science, Courant
Institute, New York University - Rob Fergus, Dept. of Computer Science, Courant Institute, New York University
47. Conclusions:
âą It's possible to predict income levels
from space
âą Underlying data can provide valuable
assistance to complex neural networks
âą Human-based empirical inquiry has legs
âą Teasing out why it knows what it knows
is interesting
48. Questions:
âą What does this thing do when you point
it at other cities?
âą What are the similarities and differences
between cities from space?
âą Can we construct a model to account for
seasonal variance?
âą Can we construct a model to account for
architectural difference?