Machine Learning in 5 minutes

Vadim Vashkelis
Sep 2, 2018 · 5 min read

Machine Learning is a real buzz-word today. At the same time, there is still a lot of confusion in terms and in general understanding of machine learning concepts. This short overview should help non-experts quickly identify the key ideas of ML in minutes.

Intro

Machine Learning, Deep Learning, AI, Data Science… I am lost, help me! Don’t panic, that is not so complicated.

What is Machine Learning

Machine Learning (ML) is all about using your “machine” (computer) to “learn” how to deal with problems without “programming”. We need machine learning when it is difficult to program the algorithm by hand.

In short, machine learning flow consists of the following step:

  1. Design (or choose) some model potentially reflecting our processing algorithm
  2. Take the data representing typical input and related output
  3. Train an our model on that data
  4. Use this trained model to make predictions on new data

OK, but how do Machines actually learn?

The same as humans — trying, failing and repeating again. During the training, we have to tune certain parameters of our model to finally get satisfying results. Normally it takes many iterations and on every iteration we automatically adjust some parameters of our model. In other words, our model remains the same, but during the training it is tuned to necessary shape.

As a human, I can train yourself or hire a trainer. Is it the same for ML?

Absolutely. In machine learning, we use a concept of supervised and unsupervised learning.

In supervised learning, we provide some pairs of inputs/outputs to help the computer (our model) train.

In unsupervised learning, we let the computer learn from the data itself without showing what is the expected output.

Source:CambridgeSpark

NB If you started to think that supervised learning is faster and more efficient, you are right.

Example of supervised learning

  • Classification: You train with the data sets representing some classes. Once trained, that machine should be able to correctly assign the classes to new, similar datasets. For example image classifications, i.e. recognition if it is cat, dog or horse.
  • Regression: Normally you have some historical data and whant machine to predict some future numbers.

Example of unsupervised learning

  • Clustering: You ask the computer to separate similar data into clusters
  • Generative Models: Once the model captures the probability distribution of your input data, it will be able to generate more data.

What is the relationship between Machine Learning, AI, Neural Networks, etc.?

Artificial Intelligence (AI) is an area of computer science researching intelligent algorithms, for example when a machine mimics cognitive functions associated with humans. Machine Learning is just a subclass of AI.

Artificial Neural Networks (NN) is a class of algorithms or architectures inspired by biological neural networks. Neural Networks simulates the way biological brains are working and learning.

Deep Learning (DL) is an architecture using the combination of the several neural networks to produce learning results comparable or superior to human.

So, machine learning is not only about Neural Networks?

Correct. There are plenty of classic algorithms that can use a machine learning approach. For example, regression.

Source: Wikipedia

Performing a regression you are trying to approximate some statistical distribution with the linear or polynomial approach. This is a pretty classic algorithm where you can use machine learning to sequentially adjust your model to achieve minimal deviation between your approximation and actual data.

What is the difference between classic algorithms and deep learning?

Each large class of machine learning algorithms can be used in certain circumstances.

Classic algorithms

If we can imagine how our model should work and there is a relatively simple correlation between input data and output features then we use classic algorithms.

Deep learning (Neural networks)

If there is a broad input data and it is not clear how input and output are linked together we normally use deep learning. The idea of deep learning is to create complicated enough neural network architecture and train it against sample input data. Since we don’t know exactly how our model will finally look like we design the architecture with many layers and nodes with the hope this model will be able to adapt to our needs.

Once designed, this model is trained and the weights of the models (signal transfer factors between the nodes) are adjusted according to certain training algorithms. Since this is a long process and many layers of the network are tuned this is called “deep learning”.

Source

Ensemble learning

Ensemble learning helps improve machine learning results by combining several models. This approach allows the production of better predictive performance compared to a single model. That is why ensemble methods placed first in many prestigious machine learning competitions.

Ensemble learning is primarily used to improve the performance of a model or reduce the likelihood of an unfortunate selection of a poor one.

Reinforcement learning

In some cases, we cannot formulate what is input and output data but have the environment able to generate a feedback. In these cases, we can use Reinforcement Learning (RL). RL does not know anything about the environment and it learns what to do by exploring the environment. It uses actions, and receive states and rewards.

Typical example when RL can be used is a learning how to play some game. The model can learn to play the game having a “reward” as a win or lose. This is the way as AlphaGo was trained.

Below is the general map of machine learning approaches to quickly identify what method can be used.

Conclusion

This article describes the basic concepts of machine learning. We did not talk about the details of individual algorithms or how deep learning is actually performed. However, this article should give a general idea of what machine learning is and and how to deal with it.

Vadim Vashkelis

Written by

Technology executive at Axinee (axinee.com), the company developing configurable and manageable AI solutions

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade