Understanding Artificial Intelligence and Machine Learning

November 15, 2021 By admin Off

Artificial Intelligence (AI), and its subsets Machine Learning and Deep Learning (DL), play a significant role in Data Science. Data Science encompasses pre-processing, analysis and visualization, as well as prediction. Let’s take a deep dive into AI, and the subsets it encompasses.

Artificial Intelligence (AI) Computer science is concerned with creating smart machines that can perform tasks that usually require human intelligence. AI can be divided into the following three types:

Artificial Narrow Intelligence

Artificial General Intelligence (AGI).

Artificial Super Intelligence (ASI).

Narrow AI, sometimes called ‘Weak AI’ is an AI that performs one task in a specific way at its best. An automated coffee machine, for example, robs coffee beans and performs a set of steps to make the coffee. AGI, also known as “Strong AI”, performs tasks that require thinking and reasoning just like humans. Google Assist, Alexa, Chatbots are all examples of Natural Language Processing (NPL) which is used. Artificial Super Intelligence is the advanced version that outperforms human abilities. It is capable of performing creative activities such as decision making, art and emotional relationships.

Let’s now look at Machine Learning. It is a subset AI that models algorithms and helps make predictions based upon complex data patterns. Machine learning is about enabling algorithms to learn and draw insights from data, make predictions, and make predictions based on previously unanalyzed data. There are many methods for machine learning.

Supervised learning (Weak AI-Task driven)

Non-supervised Learning (Strong AI, Data Driven)

Semi-supervised learning (Strong AI-cost-effective)

Enhanced machine learning Strong AI – Learn from your mistakes

Supervised machine-learning uses historical data to analyze behavior and make future forecasts. The system is composed of a defined dataset. The input and output parameters are labeled on the data. The ML algorithm analyzes the data and determines the output based on the parameters. Supervised learning is capable of performing classification and regression tasks. Image classification includes face recognition, image classification, email spam classification and identification fraud detection. Regression tasks include weather forecasting, population growth prediction and others.

Unsupervised machine learning doesn’t use any labelled or classified How to tech naija Tools need for e-learning parameters. Unsupervised machine learning focuses on uncovering hidden structures in unlabeled data, which helps systems to correctly infer a function. They employ techniques like clustering and dimensionality reduction. Clustering is the grouping of data points that have similar metrics. This is data-driven and includes movie recommendations for Netflix users, customer segmentation, purchasing habits, and other examples. Examples of dimension reduction include feature elicitation and big data visualization.

Semi-supervised machine training uses both unlabeled and labelled data to increase learning accuracy. Semi-supervised learning is a cost-effective option when labelling data proves to be costly.

When compared to unsupervised or supervised learning, reinforcement learning is quite different. It is a process that involves trial and error, finally leading to results. It is achieved using the principle of iterative improvements cycle (to learn from past mistakes). Agents can also be taught autonomous driving in simulated environments using reinforcement learning. Q-learning is one example of reinforcement learning algorithms.

Moving on to Deep Learning, is a subset in machine learning that allows you to create algorithms that use a layered structure. DL employs multiple layers to extract more advanced features from the input. In image processing, for example, the lower layers might identify edges while the higher layers could identify concepts that are relevant to humans, such as letters, digits, or faces. DL can be referred to as a deep artificial neural net. These algorithm sets are very accurate for problems such as sound recognition, image recognition and natural language processing.