Machine Learning intuitions

Written on 24 January 2025, 11:28pm

Tagged with: ,

Machine Learning is an empirical science: Try. Optimize. Repeat.

Model: is like a student; training = teaching and then is tested. Loss = how ‘wrong’ the student is when tested.

Regression (predicting values): Mean Squared Error (MSE – penalizes large errors), Mean Absolute Error (MAE – all errors are equal, less sensitive to outliers). Both MSE and MAE contain Y (the ground truth) and Y-hat (the prediction).

Classification (predicting categories): Binary cross entropy (binary classification), categorical cross entropy (multi-class category).

ML models are simpler: linear regression, decision trees, random forests.

DL models are more complex (multi-layered neural networks): input/hidden layers/output. They need high computational power.

Training process and loss: random initialisation, forward pass, loss computation, backward pass, parameter update, repeat. Student learns, teacher corrects, repeat.

CNN (convolutional): find patterns in spatial hierarchies. The filter slides (‘convolutes’) across the input. Few parameters

DNN (dense): fully connected layers, many parameters

Transformer: attention mechanisms (relationships between words)

Image generated with napkin.ai

Leave a response