deeplearning101.md

Deep Learning 101

Hands on with Python and Keras

---

About me

• Software Eng. @ trivago doing Deep Learning/Image Recognition
• GitHub Campus Expert

---

What are we going to build?

A convolutional neural network for classyfing images

---

Wait, what?

---

---

Convolutional layer

1. imagine a flashlight that is shining over the top left of the image
2. Let’s say that the light this flashlight shines covers a 5 x 5 area
3. And now, let’s imagine this flashlight sliding across all the areas of the input image.

---

---

• In CNN terminology, the flaslight is called a ‘filter‘ or ‘kernel’ or ‘feature detector’
• The matrix formed = ‘Convolved Feature’/‘Activation Map’/‘Feature Map‘
• Filters acts as feature detectors from the original input image

---

---

Plan of attack

---

1. Use pre-trained VGG16 to classify images
2. Build our own neural network
3. Use transfer learning to speedup the training with our custom classes
4. Use our saved weights

---

Pre-trained VGG16

---

Custom CNN

---

Transfer learning

---

Use own pre-trained weights

---

Find me

• Linkedin/Xing: Mikolaj Wawrzyniak
• Twitter: @spejsops
• GitHub: spejss

---