In this blog, I share my experience when I started learning ML.
Humans have long feared the rise of machines, their own creations becoming smarter and more intelligent than humans. But while artificial intelligence and machine learning are rapidly changing the world and driving the Fourth Industrial Revolution, humanity has no reason to fear.
Machine learning is a core sub-field of artificial intelligence. This puts the computer into self-learning mode without any explicit programming. As new data is input, these computers learn, grow, change, and evolve on their own.
Let’s talk about techniques related to machine learning. Machine learning techniques are mainly classified into four categories as follows:
1. Supervised learning:
Supervised learning is applicable when machines input and output accurate labeled or sample data. Correct labels are used for some labels and tags to validate model accuracy. Using prior knowledge and labeled instances, a supervised learning approach can predict future events. Before introducing an inference function to predict output values, we first explore a known training dataset. Additionally, it predicts mistakes throughout the learning process and uses algorithms to correct them.
Suppose you have a collection of photos with the keyword “cat” in the title. Using these cat pictures, a machine learning system was trained to recognize cats from other objects.
2. Unsupervised learning:
Unsupervised learning trains a computer using only a small subset of input samples or labels without knowledge of the final product. In contrast to supervised learning, the training data is unclassified and unlabeled, so the machine does not always give correct results.
Unsupervised learning is less prevalent in real-world business contexts, but it is useful for data exploration and is sometimes used to infer potential structure from unlabeled data.
Example: Suppose a computer is educated with a set of papers belonging to several categories (types A, B, and C) and should be classified according to their relevance. Computers can classify these datasets into Type A, Type B, and Type C categories, given only an input sample or no output.
3. Reinforcement learning:
A feedback-based machine learning technique is reinforcement learning. This kind of learning requires an agent (a computer program) to explore its surroundings, take action, and be rewarded as feedback for its activity. They receive a positive reward for every good deed and a negative reward for every bad deed. The goal of reinforcement learning agents is to maximize good results. Since there is no labeled data, agents can only learn from experience.
4. Semi-supervised learning:
Semi-supervised learning, a technique used in both supervised and unsupervised learning, falls somewhere in between. Both datasets with few labels and datasets containing unlabeled data can be manipulated. However, the data are usually unlabeled. Labels are expensive, but they may not be needed for your business goals, so they also reduce the cost of machine learning models. It also improves the performance and accuracy of machine learning models.
Semi-supervised learning allows data scientists to overcome the limitations of supervised and unsupervised learning. Important applications of semi-supervised learning include speech analysis, online content classification, protein sequence classification, text document classifiers, and more.
This topic may seem difficult, but it gets easier over time as interest in the field grows. I started learning ML by watching some YouTube tutorials and reading about the topic.
First of all, it is important to learn a programming language. For ML, you can learn Python or R language. Python’s low complexity makes it an easy and suitable language for beginners.
I learned Python from learnpython.org. It’s free to use and beginner-friendly.
Why do we need math here? Some machine learning algorithms are built using mathematical concepts. For machine learning, you don’t need to be a math master, just learn a few basic concepts. Basic topics such as linear algebra, statistics, probability, and calculus should be covered.
The Python language includes several built-in libraries used in machine learning. Several libraries such as NumPy, Matplotlib, Pandas and Seaborn are used. NumPy is used for processing mathematical functions and arrays, Pandas is used for data set manipulation, and Matplotlib and Seaborn are used for data visualization.
Learning these libraries will be very helpful when building your ML projects.
One of the most important steps is data preparation. Machine learning involves working with large datasets. Before we can implement ML algorithms on these datasets, we need to prepare them. The dataset may contain null values or duplicate values.
- Machine learning algorithms:
It can be categorized into three types: supervised learning, unsupervised learning, and reinforcement learning. These algorithms include linear regression, logistic regression, K nearest neighbors, support vector machines, random forests, decision trees, and more. It is very important to learn the concept and coding part of each algorithm.
Practice is the only key to a solid understanding of machine learning concepts. Participating in hackathons, building projects, and participating in Kaggle contests can go a long way.
Author: Saloni Chowdhury
