-
Applications
-
Images
- Image classification
- Image semantic segmentation
- Image retrieval
- Object detection
-
Language
- Text classification
-
Software mining
- Software flaw detection
-
History
- Receptive field in a single neuron, 1959
- Neocognitron, about 1980
- Gradient-based CNN for hand-written character recognition, 1998
- AlexNet in ImageNet competition, 2012
-
Hierarchical structure
-
Input: raw data
-
Types
- RGB images
- Raw audio data
-
Properties
-
3D tensor
- H-row
- W-height
- 3-channels
-
Feed-forward
-
Operations to abstract each layer
- Convolution
- Pooling
- Non-linear activation function
- Or combinations
- Last layers: objective function
-
Backpropogation
-
Loss
-
Loss functions
- Mean Square Error
- Cross Entropy Error
- Loss in precious layers