Home » What are the Principles of Machine Learning

What are the Principles of Machine Learning

Artificial Intelligence and Machine Learning based advancements are increasingly making our daily lives more efficient and convenient. AI and ML are the technologies driving many of these programs and services that help us do daily tasks such as connecting with friends on Social Networks, using an Email program, or using a Cab service. Most of us have been using AI daily for many years now without fully being aware of it.

Digital Assistants, Navigation through Maps, Parking Vehicles, and Self-Driving vehicles, Smart E-mail communication, Predictive Web search, Enhanced Online Stores and Services, Social Media Apps, Commercial Airlines, Medical diagnoses, and health care- are some of the common examples of AI and ML being applied in several real-life situations. Therefore, Students and working people interested in a career in ML now have several options for the study program, and the Machine Learning certification program can be an attractive proposition. But let’s get an idea of what’s Machine Learning.

Machine Learning (ML)

A division of artificial intelligence, Machine Learning, understands that computer systems can make decisions using algorithms and statistical models, which helps analyze and draw inferences from data and data patterns, with minimal human intervention.

Types of Machine Learning

There are three broad types of machine learning:

  1. a) Supervised learning
  2. b) Unsupervised learning
  3. c) Reinforcement learning

Supervised learning

The supervised learning approach is very similar to human beings learning under the supervision of a teacher.

This algorithm learns from the pattern of data and the associated response, which it uses to predict the correct response when input with a new data set.

Examples:

E-mail Spam Classification – The algorithm analyses data patterns to classify spam from others’ mails, using historical spam and non-spam e-mails as input.

Stock Price Prediction – The algorithm predicts the new future price of a stock, based on proper regression analysis of historical business and market data,

Unsupervised Learning

In Unsupervised Learning, learning occurs when an algorithm learns from examples and determines the data patterns.

Example: In retail analytics, customers are usually clustered together based on their purchase preferences and other behaviors.

Reinforcement Learning

Reinforcement Learning is different from the above two types of learning, wherein the learning process of the model is reinforced. It is a reward and punishment-based learning system in which a machine learns a series of actions to perform a task. Each action of this algorithm is connected with either a positive (reward) or negative (punishment).

Example: Computers learn to play video games by themselves. The algorithm keeps interacting with the game environment through a series of actions, giving a reward or punishment based on the nature of the action taken.

Key Elements of Machine Learning

Machine learning algorithm has three components:

  • Representation: how to represent knowledge.

Examples are decision trees, sets of rules, instances, graphical models, neural networks, support vector machines, model ensembles.

  • Evaluation: The way to evaluate candidate programs (hypotheses).

Examples are accuracy, prediction and recall, squared error, likelihood, posterior probability, cost, margin, entropy k-L divergence.

  • Optimization: The way candidate programs are generated in the search process.

Examples are combinatorial optimization, convex optimization, and constrained optimization.

All machine learning algorithms are combinations of these three components.

Principles of Machine Learning

Following are the key principles of ML:

  1. Human augmentation
  2. Bias evaluation
  3. Explainability by justification
  4. Reproducible operations
  5. Displacement strategy
  6. Practical accuracy
  7. Trust by the privacy
  8. Data risk awareness

1. Human augmentation

Commitment to study the impact of wrong predictions and design review systems and processes with human input looped in when reasonable.

When you introduce automation through machine learning systems, it’s easy to forget the impact of wrong predictions on the end-to-end automation process.

Enabling Subject domain experts as human reviewers in the loop at the end of ML systems can have significant benefits.

2. Bias evaluation

Commitment to developing processes that allow understanding, documentation, and monitoring biases in development and production.

3. Explainability by justification

Commitment to developing tools and processes to help sustain improvements in transparency and explainability of machine learning models.

In deep learning, technologists throw large amounts of data into complex ML conduits hoping something will work, without understanding how the pipelines work internally. However, technologists should invest reasonable efforts to continuously improve tools and processes that allow them to explain results based on features and models chosen.

4. Reproducible operations

Commitment to develop the required infrastructure and enable a reasonable level of reproducibility across the operations of ML Systems.

Typically, production machine learning systems cannot diagnose or respond effectively when something bad happens with a model. There are various tools and best practices that you must follow for machine learning operations.

5. Displacement strategy

Commitment to identify and document information about business change processes to help mitigate the impact of automation on workers’ livelihood.

6. Practical Accuracy

Commitment to developing processes that ensure accuracy and regulation of cost functions aligned to domain-specific applications. When you build systems that learn from data, it is important to understand the underlying means to ensure accuracy.

7. Trust by Privacy

Commitment to build and communicate processes that protect, secure and handle data with stakeholders interacting with the system directly or indirectly.

8. Data risk awareness

Commitment to develop and improve reasonable processes and infrastructure to ensure data and model security during the development of machine learning systems.

Conclusion:

The above basics of machine learning can help a budding ML enthusiast get an overview of machine learning and appreciate how ML works behind the scenes in several daily life applications. Those seeking a career in ML have several online courses available that they can choose as per their interests. These include:

  • Web search: ranking page based on what you are most likely to click on.
  • Finance: Targeting particular credit cardholders with offers.
  • E-commerce: Predicting customer preferences.
  • Space exploration: Space probing satellites and Radio astronomy.
  • Robotics: To handle uncertainty in new scenarios. E.g., Self-driving cars.
  • Information extraction: Conduct automated surveys across the web.
  • Social networks: Machine learning extracts value from data on relationships and preferences in social networks.
  • Debugging: Running Debugging process in computer science or Codes.

Image Credit – freepik.com

Leave a Comment