What is AI?
Artificial Intelligence is the ability to make machines intelligent enough to perform tasks that a human would otherwise be able to perform naturally, such as visual perception, speech recognition, decision-making based on self- learning, and language translation etc.
Machine learning and deep learning are different branches of AI that enable computer systems to self-learn and adapt with the assistance of intelligent data analysis and powerful learning algorithms.
What is Machine Learning?
Machine learning is the science of training a machine or software or computer system to learn from experience, without the need to explicitly program it.
There are multiple Machine Learning algorithms which interpret data and detect hidden patterns to derive a logical or predictive model.
These algorithms progressively adapt to improve their performance based on increasing sample data size.
Machine Learning algorithms can be categorized according to the techniques applied - Supervised, Unsupervised and Re-inforcement.
Supervised learning algorithms: These algorithms handle labeled datasets, and are trained to identify input data attributes that map to a known set of outputs. Linear Regression, Random Forst and Support Vector Machines are different algorithms based on supervised learning.
Classification is one such technique input data is categorized to produce discrete responses,
e.g. email spam detection (identifying if an email is genuine or spam), medical imaging to identify critical conditions (e.g. if a tumor is to be considered cancerous or benign), software based on voice-commands (s/as Cortana), and to check creditworthiness based on clients' credit scoring.
Regression is another such technique, where the input data is dynamic and frequently susceptible to change,
e.g. electricity load forecasting (while considering variations in power consumption), algorithmic trading in the finance industry (while taking into account the constant fluctuations in stock prices) and in weather applications (to predict a 'sunny' day if the temperature is above a threshold or a 'cloudy' day if the humidity is high).
Unsupervised learning algorithms: work with datasets that are unlabeled and have no output associations. These algorithms self-train based on identifying hidden patterns and similar behavioral characteristics in the unlabeled data sets. K-Means, Apriori and Hierarchical Clustering are such Unsupervised learning algorithms.
Clustering: analyses unlabeled input data to find logical groupings or clusters, and break them into different classes, with the goal to achieve high intra-class similarity and low inter-class similarity.
e.g. segmentation of customers (high net worthiness) in the banking sector, for statistical data analysis such as gene sequencing analysis, market research and object recognition (based on characteristics of different objects captured and processed).
- Collaborative Filtering: recommendation of new products to customers based on past purchases as well as purchasing patterns of other customers (ever noticed the section 'Customers Who Bought This Item Also Bought' on Amazon? This feature was introduced using Amazon's proprietary algorithm based on 'affinity analysis' that works on the principle of correlation between continuous variables.)
- Re-inforcement learning algorithms: These algorithms analyze the input data and train themselves in two ways -
- Exploration: by making new calculative choices by way of trial and error, and
- Exploitation: by learning from its environment using a reward-based system. The following illustration is one such example of a reward-based system
What is Deep Learning?
Deep learning, also known as deep structured learning or hierarchical learning, is a sub-field of machine learning that breaks down tasks in ways to make better and practical application of AI possible in real world scenarios. Deep learning algorithms are inspired by the structural and functional design of the human brain called Artificial Neural Networks (ANNs).
Artificial Neural Networks (ANNs) are inspired by the interconnections of the neurons in the brain. A neuron (or a nerve cell) is the basic working unit of the nervous system, which takes up, processes and transmits information to other neurons. A neural network is a parallel distributed processor that stores knowledge from experience and makes it available to use for future predictions.
ANNs work on discrete layers, connections between the layers and data propagation. For example, when data is inputted into the first layer of the neural network for processing, the neurons in the first layer will process and pass on the output into the second layer. The second layer of neurons will then perform its task and pass on the output to the next layer, and so on till the final output is produced. Each neuron will assign a weighting to its input - how correct or incorrect it is relative to the task being performed. The correctness of the final output is determined by the total of the weightings.
An ANN is similar to the human brain in two ways - it acquires knowledge through learning and works upon inter-neuron connection strengths. The most important design considerations in an artificial neural network are finding the right number of hidden layers, the number of neurons per layer, and the number of input and output nodes. This is more or less decided on a trial and error approach - too few neurons could result in large errors, and too many could result in over-training.
ANNs learn from examples and generate mapping relationships between inputs and outputs. In order to generate this self-organizing map, the ANN is trained, tested and validated with Supervised and Unsupervised learning techniques.
One step into Artificial Intelligence
Machine and deep learning algorithms find natural patterns in massive amounts of data to find insights and help make predictive and critical decisions. Advanced techniques which help industries include image processing and computer vision, for face recognition, speech recognition, object detection and motion detection.
Decisions are being made on a daily basis in fields such as:
Medical Research & Diagnosis – such as computational biology for tumor detection, DNA sequencing and drug discovery.
Financial Computations – for credit scoring, building algorithmic trading models and fraud detection.
Retail Experience – retailers use it to gain insight into their customers’ purchasing behaviour.
Energy Forecasting – energy production and consumption, for price and load forecasting.
Social Media – media sites such as Google, Facebook sifting through millions of options to increase traffic or make purchase and product recommendations and generate ad revenue.
Advanced Transport – manufacturing and predictive maintenance of automotive and aerospace technologies.
Natural Language Processing (NLP) – where machines are able to comprehend how humans organize their thoughts, feelings, language and behaviour.
Robotics – physical bots automated for manual but intelligent labour such as the use of robotic arms in spacecraft.
The Internet of Things (IoT) – inter-connecting smart devices (e.g. home appliances, monitoring and maintenance systems etc.) for a smarter and more efficient world.
Organizational & Corporate Efficiency – self-learning software that improves employee management and performance, Customer Relationship Management (CRM) analytics, sales and strategic planning.