What if you could run a deep learning model in reverse? Train up an MNIST classifier, and get the "perfect" image representation of the number 5?
<h1>Questions</h1>
<ul>
<li>You can't inverse the ReLu function, right?</li>
</ul>
<h2>Inversing ReLu</h2>
If using tanh instead of relu, the inverse is
<math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mfrac><mn>1</mn><mn>2</mn></mfrac><mo>∗</mo><msub><mrow><mi>log</mi><mo>⁡</mo></mrow><mn>2</mn></msub><mo stretchy="false">(</mo><mfrac><mrow><mo stretchy="false">(</mo><mn>1</mn><mo>+</mo><mi>x</mi><mo stretchy="false">)</mo></mrow><mrow><mo stretchy="false">(</mo><mn>1</mn><mo>−</mo><mi>x</mi><mo stretchy="false">)</mo></mrow></mfrac><mo stretchy="false">)</mo></mrow><annotation encoding="application/x-tex">\frac{1}{2}*\log_2(\frac{(1+x)}{(1-x)})</annotation></semantics></math>21​∗log2​((1−x)(1+x)​)


What if you could run a deep learning model in reverse? Train up an MNIST classifier, and get the "perfect" image representation of the number 5?

# Questions

- You can't inverse the ReLu function, right?

## Inversing ReLu

If using tanh instead of relu, the inverse is
$\frac{1}{2}*\log_2(\frac{(1+x)}{(1-x)})$


Reversing Neural Networks

Questions

Inversing ReLu