Machine Learning index
@index
Machine learning courses
Neural nets
Tasks:
- Segmentation
- Object Detection
- Classification
- Regression
Types of Neural Networks
- LSTM
- CNN
- Fully convolutional networks (mainly for segmentation)
Techniques:
- Stochastic Gradient Descent
- Weight initialization
- Regularization in CNNs
- Batch normalization
- Residual skip connections
- Backpropagation
- Why to choose odd-sized convolution kernels
- Increase the amount of channels for deeper convolution layers
- Non-max Suppression Algorithm for Bounding Boxes
Other relevant concepts:
Normalization and regularization
Classification
Inference
Sampling techniques
Clustering
- Elbow method for picking an appropriate amount of clusters in an unsupervised manner
NLP
See index Natural Language Processing (NLP)
Loss functions
Evaluation
Classification:
TODO
Regression:
TODO
Object detection:
Jaccard Index: intersection over union * Used for segmentation tasks. * Useful for bounding boxes in object detection
Forecasting
Explainability
Emsemble learning
Two main approaches: