Machine Learning

Machine Learning is the study of Artificial Intelligence algorithms that learn to perform tasks using data which is used to create a data model in a training process, instead of using explicitly coded logic. The objective in machine learning is to capture regularity in data to make predictions and inferences.

Components of Machine Learning:

Re-Iterate untile metrics are satisfactory
Data Preparation
Data Preprocessing & Wrangling
Feature Extraction & Engineering
Feature Scaling & Selection
Data Retrieval
DataSet
Modeling
Model Evaluation & Tuning
Deployement & Monitoring
Machine Learning Algorithm
Machine Learning Workflow

Different Classes(Paradigms) of Machine Learning, with first three being being the major ones:

  • Supervised Machine Learning: In Supervised Machine Learning, the dataset already has labels for prediction which is used to train the model for new variables by associating features with corresponding outputs.

    • 📝Uses Categorical Target Variables in Clustering and Continuous Target Variables in Regression tasks.
  • Unsupervised Machine Learning: In Unsupervised Machine Learning however, instead of performing predictions using already existing labels, the training is used for discovering hidden patterns and structures in unlabeled data.

    • 📝In Unsupervised Machine Learning, Target Variables are not available.
  • Reinforcement Learning: In Reinforcement Learning, the algorithm earns rewards or penalties based for it's actions to reinforce good behavior and become become optimized.

    • 📝Can be used for control tasks without target variables, or use Categorical Target Variables in Classification.
  • Semi-Supervised Learning

  • Self-supervised Learning

    Other less popular Machine Learning paradigms and methods:

    • Meta learning: Automatic learning algorithms are applied to metadata about machine learning experiments.
    • Quantum machine learning: Use quantum algorithms within machine learning programs.
    • Online learning: Use sequence of data to improve prediction.
    • Batch learning: Use entire dataset to improve prediction

Types of Machine Learning Tasks:


Approaches in building Machine Learning models:


Factors effecting the performance(accuracy) of a model:

Info

Often Model Compression Techniques are utilized to reduce the size of model at the cost of small drop in accuracy. This techniques can also lead to faster training time, lower resource consumption, and faster inference time.


Machine Learning Vs Traditional Algorithms

Traditional algorithms: All instructions are explicit and only input parameters effects the function.

Machine Learning: Machine Learning functions use a data model generated during training process. Model is trained from dataset and is effected by hyperparameters evaluation process.


Types of Machine Learning models:

  • Generative Models: They model distribution of the classes within a dataset to find joint probability(Where given input feature coexists with the desired output) to generate new data such as images, text, code, and audio(music, or speech). They use joint probability distribution: .
  • Discriminative Models: They model boundary between classes within a dataset to find the decision boundary(Where boundary of classes are distinguished or the output label of a classifier is ambiguous) and discriminate based on it's training data to make predictions or classifications. They use Conditional Probability: .
  • Parametric Models
  • Non-Parametric Models
  • Tree-Based Models
  • Non-Tree-Based Models

Differences between Deep Learning and traditional Machine Learning:

  • In Deep Learning the task of Feature Extraction and preparing model for specific patterns is performed by learning model, instead of a Data Scientist or AI Developer.
  • Deep Learning model is more complex and can perform general tasks much better. in contrast traditional Machine Learning is domain specific.

Algorithms


Learning Material:

Resources:

References: