MACHINE LEARNING- A walkthrough

Machine learning is an application that gives systems the ability to automatically learn and improve from experiences. It involves computers discovering how they can perform tasks without being explicitly programmed to do so. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves so that they carry out certain tasks.

The learning begins with observations or data, direct experience, or instruction, to look for patterns to make better decisions in the future based on the examples provided. For simple tasks assigned to computers, we can program algorithms telling the machine mode of execution of all steps required, on the computer's part, no learning is required. For more advanced tasks, it can be challenging to manually create the needed algorithms.

The primary goal is to program computers to learn automatically and adjust actions accordingly. In practice, it is more effective to train the machine to develop its own algorithm, rather than having human programmers. For example, by using the classic algorithms of machine learning, the text is considered as a sequence of keywords; instead, an approach based on semantic analysis mimics the human ability to understand the meaning of a text.

Machine learning uses various approaches to teach computers to perform tasks even where no fully satisfactory algorithm is available. In cases where vast numbers of potential answers exist, one approach is to tell the system some of the correct answers. This is then; used as training data for the computer to improve the algorithm it uses.

A LITTLE BIT OF HISTORY FOR PERSPECTIVE

The named Machine Learning was given by 1959 by an American IBMer and pioneer in the field of computer gaming and artificial intelligence, Arthur Samuel. There has been growing interest related to pattern recognition in the 1970s using this method, and then in 1981, a report was given on using teaching strategies so that a neural network learns to recognize 26 letters, 10 digits, and 4 special symbols from a computer terminal.

Tom M. Mitchell a more formal definition of machine learning field: "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." ML does tasks fundamentally operational in nature rather than cognitive.

KEY ELEMENTS OF MACHINE LEARNING

For dThere are millions of new machine learning algorithms developed each year. Every machine learning algorithm has three components:

Representation:   How to represent knowledge. Examples include decision trees, sets of rules, instances, graphical models, neural networks to name a few

Evaluation:  The way to evaluate the Hypotheses. Examples include prediction and recall, squared error, likelihood, posterior probability, cost, margin to name some.

Optimization:   The way the hypothesis is generated, is known as the search process. For example, combinatorial optimization, convex optimization, constrained optimization. All machine learning algorithms are a framework for understanding all algorithms using combinations of these three components.

TYPES OF MACHINE LEARNING

There are four types of machine learning:

Supervised learning is Training data that includes desired outputs. This approach teaches machines by example, learning is supervised.

Unsupervised learning involves training data that does not include the desired outputs. Unsupervised learning tasks algorithms with identifying patterns in data, trying to spot similarities and variances and categories into segments. An example is clustering. Its quality of learning is not guaranteed.

Semi-supervised learning: This approach mixes supervised and unsupervised learning. The technique relies upon using a small amount of labelled data and a large amount of unlabelled data to train systems. The labelled data is used to partially train and then that partially trained model is used to label the unlabelled data, a process called pseudo-labelling. The model is then trained on the resulting mix of the labelled and pseudo-labelled data.

Reinforcement learning: Rewards from a sequence of actions. Reinforcement learning is to think about how a complete novice, eventually, by looking at the relationship between the buttons they press and their in-game score, will learn and as a result, their performance will get better and better.

Supervised learning is the most mature & Learning with supervision is much easier than learning without supervision. It begins with training a machine-learning model, a mathematical function that is capable of self -modifying until it can make accurate predictions with new fresh data.

Frankly, this basic process of finding the pattern and apply the pattern much runs the world. Thanks to an invention in 1986, by Geoffrey Hinton, known as the father of deep learning.

UNDERSTANDING DEEP LEARNING

Deep learning is machine learning on steroids: it uses a technique that enhances the machines’ ability to spot and amplify even the smallest patterns. This technique is called a deep neural network, as it has many, many layers of simple computational nodes that work together to process and regress data and to deliver a result in the form of the prediction.

There are two types of machine learning software on the market.

Open source and proprietary software. Popular free and open source include Caffe, CNTK, deeplearning4j, deepseed, ELKI, infer.net, keras, mahout, mallet, ML.net, mlpack, MXnet, Neural lab, Octave to name a few.

Amongst the more established proprietary ones are Amazon machine learning, Angoss knowledge studio, Azure machine learning, IBM data science experience, Google prediction API, IBM knowledge modeller, KXEN modeler, LION server, Mathematic, Matlab, Microsoft Azure, Neural designer, Neural solutions, Oracle data mining, SAS enterprise miner, SequenceL and STATISTICA data miner.

LIMITATIONS OF MACHINE LEARNING

Machine learning algorithms are also limited by a few factors including Bias which the system can pick up during unsupervised learning and then go undetected. The model assessments are also subjective to the individual’s interpretation of the outcome, this can also occur due to the data splitting methods where some of the data is used to predict the rest. And ethics, the eventual result as shown to the users and as used is all dependent on what is shown and how it is presented to the users. People do tend to bring in personal experience and biases into decision making which eventually do seep into the system.