top of page

What Is Machine Learning (ML)



Machine Learning (ML) is a fascinating field of AI that can allow us in many ways to understand the principles of computer intelligence. As a field of applied statistics, ML gives us insight into a different process of learning in itself, such as learning from data or answers instead of by defined rules. There is much to discover about ML, and the potential for technological growth in this area of AI is rapidly advancing. For a basic definition:

Machine Learning (ML) is a subset of AI in which machines acquire their own knowledge by extracting patterns from data without being explicitly programmed (DL). With a ML algorithm, a computer learns from data to make intelligent decisions, solve real world problems, perform an action or numerous “human-like” tasks.

IMPORTANT TERMS

Features- each piece of representative data that is given to the computer to learn from.

Example- collection of features that have been quantitatively measured from some object or event that we want the machine learning system to process. (DL)

Dataset- collection of many examples. Examples are also called data points. An ML algorithm learns from the dataset it is given.

Hyperparameters- settings of ML algorithms. Hyperparameters must be determined outside of the learning algorithm.

Model- The representation (internal model) of a phenomenon that a Machine Learning algorithm has learned. It learns this from the data it is shown during training. The model is the output you get after training an algorithm. (Gavin Edwards “Machine Learning | An Introduction). Gavin Edwards "Machine Learning | An Introduction" https://towardsdatascience.com/machine-learning-an-introduction-23b84d51e6d0#0ea7

THE IMPORTANCE OF DATA

The performance of Machine Learning depends on the representation of data that it is given to learn from. Each piece of representative data that is fed to the computer is called a “feature”, and must be relevant to what the algorithm is trying to accomplish. It can be quite complex and tedious for AI scientists to decide what information is relevant to give the machine, but ML data must be the following:

  • An Adequate Data Amount

  • An Adequate Data Depth

  • Well-balanced Data

  • Highly representative and Unbiased Data

  • Complete Data

  • Clean Data

Depending on the quality of data, the accuracy (proportion of examples for which the model produces the correct output DL) is likely to increase.


THE LEARNING PROCESS

The way that a machine learns is often complicated to break down into simple terms, but Tom Mitchell (1997) states:

“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.” -Tom Mitchell, AI Computer Scientist & Professor at Carnegie Mellon University

What this basically means is if there is a task (T) that we want the machine to solve, we design the measure (P) to measure its performance. The experience (E) is what kind of experience we allow the machine to have during the learning process.

According to Yufeng G’s “The 7 Steps Of Machine Learning”, the basic process for machine learning is thus:

  1. Gathering Data

  2. Data Preparation

  3. Choosing The Model

  4. Training

  5. Evaluation

  6. Tuning

  7. Prediction

For more detail on each step of Machine Learning, use the following link to the article source:


TYPES OF MACHINE LEARNING PROBLEMS



There are many types of experiences that we can allow a machine to have while learning, but these are the most common:


Unsupervised Learning- learning algorithm where the machine observes several examples of a random vector x, and then attempts to learn the probability distribution of the dataset or other interesting features. This algorithm is only provided with inputs in order to learn how to find hidden patterns within the data. There is no instructor or guide to teach the algorithm how to make sense of the data it is shown.


Supervised Learning- learning algorithm where the machine observes random values of x and y, and then learns to predict target y from x. An instructor or teacher is provided to teach the machine the mapping relationships between inputs and outputs. In other words, the instructor shows the ML system what to do with the data it is given, and the machine then copies the approach to make accurate estimates p(y | x).


Reinforcement Learning- learning algorithm that interacts with an environment, in which there is a feedback loop between the learning system and its experiences to maximize the reward received at the end of the process. This type of learning is most similar to the way that humans think and function, in which the algorithm learns by itself through trial and error what to do.


TYPES OF MACHINE LEARNING TASKS & EXAMPLES

Based on the kind of problem the machine is supposed to solve, there are many different approaches to design ML algorithms that will succeed in solving the task.

Classification- the computer program is asked to specify which of k categories some input belongs to based on one or more independent variables. It is the process of assigning a “class label” to a particular item, determining which class the item belongs to. This is usually done by grouping similar data points together to classify them into different sections. Examples:

  • Spam Detection

  • Churn Prediction

  • Sentiment Analysis

  • Facial Detection

Regression- the computer program is asked to predict a numerical value given some input. The algorithm takes the input data, passes it through a predictive model, and generates a number value (such as a stock closing price). Examples:

  • Price prediction

  • Stock prediction

  • Revenue and growth forecasting

  • Credit/Loan Default Risk

Transcription- the ML system is asked to observe an unstructured representation of data and transcribe the information into a discrete textual form. Examples:

  • Speech Recognition

  • Optical Character Recognition (computer is shown an image with text and must return as a sequence of characters).

Machine Translation- the computer program is given a sequence of symbols of some language and must convert it to a sequence of symbols in another language. Examples:

  • English to French (Google Translate)

Anomaly Detection- the computer program sifts through data objects and flags some of them as unusual or abnormal. Examples can be applied to several areas of cybersecurity and network security:

  • Credit Card Fraud

  • Malware Detection

  • Phishing Attack Detection

CONCLUSION

To recap, machine learning is a subset of AI in which machines are able to learn from data patterns without being explicitly programmed to do so. The capabilities of ML are astounding in the fact that machines train themselves instead of being led step by step by human example. The applications of ML extend far beyond the mere tasks and examples cited in this post, but below are more interesting projects that are currently in effect. Be sure to review them!

SOURCES


bottom of page