Is Your Machine Really Learning & Helping You Improve Your Process?

Machines learn with the help of 'Machine Learning.' Machine learning enables computers to take care of tasks that were earlier not possible without human help. Machine learning is a branch of artificial intelligence. In the modern world, machine learning is everywhere. How does Facebook suggest stories? It's because of Machine Learning. With Machine learning, computers improve and learn from experience. They work on predictions and perform tasks automatically. 


In the process of Machine Learning, we teach computer systems to make accurate predictions with fed data.

Predictions can be anything- it can be a self-driving car that can spot people crossing the road, identifying whether a fruit in the picture is apple and banana. It is because of Machine Learning that the word "book" in a sentence means a reservation or a paperback. The interesting thing is that human developers don't need to write a code to help computers distinguish between an apple and a banana and provide the right answer. 

A machine-learning model is taught how to distinguish between the two by feeding a large amount of data, for example, a vast amount of images containing an apple and a banana.

Machine Learning is possible, with a huge amount of data.


We see Machine Learning everywhere around us. It is all thanks to Machine Learning that we get recommendations on what we might be interested in watching on Netflix, what we want to buy on Amazon, etc.

A variety of Machine-Learning systems are used by Google to understand the language in your searches. It provides personalized results so that those who are searching for a paperback "book" don't get results about booking hotels. Machine-Learning is used to recognize spam in your Gmail so that you don't get disturbed by messages that are not relevant to you. Similarly, Gmail's spam and phishing-recognition systems use machine-learning trained models to keep your inbox clear of rogue messages.

Virtual assistants like Apple's Siri and Amazon's Alexa, Microsoft's Cortana all understand natural language and support voice recognition with the power of Machine Learning. 

Apart from that, Machine Learning is finding its way into every industry. It is used in drones and delivery robots, synthesis for chatbots and service robots, computer vision for driverless cars, speech and language recognition, and facial recognition for surveillance in countries like China. Not only that, Machine Learning is helping radiologists to identify tumors in x-rays. It also helps researchers to get to know about genetic sequences concerning diseases and getting to know about molecules that could pave the path for more effective drugs in healthcare. It will greatly help in predictive maintenance on infrastructure by analyzing IoT sensor data, and the list goes on and on.


In supervised learning, machines are taught by example. Systems go through training for supervised learning, and during the training, they are exposed to a huge amount of labeled data. There are so many advantages of Machine Learning. Labeled data, for instance, handwritten numbers; computer systems learn to recognize which number they correspond to. In supervised learning, a huge amount of examples are provided to help computer systems learn to recognize pixels and shapes equivalent to each number. But, it is not a cakewalk to train these systems as a huge amount of labeled data is the first requirement and some systems need to be provided with millions of examples. 

Vast datasets, such as, ImageNet, that has over 14 million categorized images are required for the task. Google's open images dataset and YouTube-8M are some of the examples. Facebook has compiled 3.5 billion images that were available publicly on Instagram by using hashtags attached to the images as labels. They yielded an accuracy of 85.4 on ImageNet's benchmark, which is a record in itself.

Crowdworking services, like Amazon Mechanical Turk, help in labeling the datasets. Crowdworking services provide a large pool of low-cost labor spread across the globe. ImageNet had 50,000 people working under the company, and the majority of these were recruited by Amazon Mechanical Turk. When we see the approach of Facebook, it's different. Facebook eliminated the need for manual labeling by using publicly available data.


In the machine-learning model, a mathematical function continuously modifies how it functions until it can make correct predictions when provided with fresh data.

For training, you have to decide which data to capture and what features of data are crucial.  

For the understanding of features, let’s take an example of Explainer by Google. Here, a Machine Learning model is trained to identify the difference between beer and wine by keeping in view the alcoholic volume (ABV) and the color of beverages. A spectrometer is used to measure their color and a hydrometer to measure the alcohol content.

The data should have an approximately equal number of beer and wine. 

70% of the gathered data is then split for training and the remaining 30% for evaluation. Evaluation data helps in testing how well the trained model can perform on real-world data. 

Before starting with the training, there are steps like data-preparation, checking the duplicate content, error correction, and normalization. After that, a machine-learning model is chosen out of the so many available. The next step will be choosing an appropriate from the wide variety available. Some machine-learning models are best for handling numerical data, while others are great for handling images.


In the training process, the machine-learning model automatically tweaks how it functions until it learns to make accurate predictions from data. Take the example of beer and wine above; the learning process continues until it learns to label a drink as beer or wine when the model is given a drink's ABV and color.

Take the example of linear regression with gradient descent to learn about the training process. For example, we need to find out how many ice creams can be sold based on the outside temperature. To do this, past data containing ice cream sales and the outside temperature is plotted against each other on a scatter graph to create a scattering of discrete points. 

Final Words

With the advancements in technology, machines are really learning and helping in the betterment of processes. In the article, we have described Machine Learning and how machines are trained. Every sector is leveraging the power of Machine Learning. 

Post a Comment