Layman's Terms: Machine Learning
Target Audience? For those of you out there with little to no previous knowledge or experience of Machine Learning (ML), but looking for a non-technical explanation of what exactly it is.
The Goal? Provide you with, in layman’s terms, a basic understanding of Machine Learning (ML).
Prior knowledge? NONE! This blog is intended for all readers, regardless of background or technical expertise….
So, isn’t machine learning all about robots taking over the world?......
NOPE - let’s bury this myth from the get-go…PHEW!! 😊
While many of your favourite films over the years such as Terminator and The Blade Runner have depicted robots developing emotions and self-awareness beyond the control of humans, don’t expect anything similar to happen in reality. While the thoughts of self-controlling robots shooting laser beams from their eyes sounds exciting/terrifying, machine learning instead focuses on:
1. Detection (interpreting the present)
2. Prediction (learning from the present and predicting future outcomes)
How boring, right?
Well, no not really. Before we tell you why it’s not as boring as you might think (albeit with no real-life Terminator), let’s examine what ML concept actually is first.
To do that, we must first of all understand Artificial Intelligence (AI)
The modern understanding of AI/ML is that AI is the study of intelligent, complex systems. They can be developed by either getting engineers to work on them (old school AI), or by having the system learn from examples (Machine Learning)
ML is a sub-category of AI, and can be described simply as machines imitating and adapting human like behaviour. The ML process is essentially humans trying to teach machines to learn from experience, through the use of algorithms. These algorithms learn iteratively, meaning that their performance and accuracy improves as the number of data samples used to train the model increases.
ML was previously viewed as a branch on the AI tree but due to the increasing complexity of modern AI, it is no longer a sub-domain. Instead, ML is transitioning to the go-to approach for all of the nodes in the AI tree. Since modern AI is so complex, almost all AI systems developed nowadays employ ML.
ML has advanced hugely over the last 10 years, with algorithms now achieving human like performance or better on a range of tasks including facial and object recognition. The algorithms used depend upon the type of learning required, with there been three main types of learning:
So, why use machine learning?
Organisations nowadays are creating and collecting more data than ever. To gain a better understanding of what ML can achieve, let’s examine how some organisations are using ML to create more innovate solutions using data:
Pinterest uses machine learning algorithms for various aspects of their business including spam moderation, content discovery and advertising monetization.
Edgecase leverages algorithms to assist online shoppers select outfits based on analysing behaviours, in an attempt to make casual browsing online more rewarding.
Netflix uses algorithms to recommend films to users, based on their previous selections.
Google’s DeepMind platform is capable of mimicking the thought process of human brains, with the platform based used within healthcare applications to assist in reducing the time it takes to plan treatments and to help diagnose ailments.
and most importantly…..
Getvisibility provides a product that utilises the latest technology in ML for data classification to give companies visibility, control and a strong understanding of their data as it is been created.
Okay, it’s starting to make more sense…
As the previous examples show, most industries working with large amounts of data have recognised the value of ML technology, and appreciate the competitive advantage it can create for them. We now see many organisations scrambling to integrate machine learning into their functionality and offerings, and with that, the demand for data scientists and ML experts has grown exponentially. Well – trained machines now possess the capabilities to do high-frequency repetitive tasks with high accuracy without getting bored. In fact, we experience some facet of ML each and every day, even though we may not even realise it:
Smartphones detecting faces while taking photos or unlocking themselves
Social media sites recommending friends and ads you might be interested in
Amazon recommending products based on your browsing history
Banks using ML to detect fraudulent activity in real time
Wow, I never realised! What about the ML process?
The ML process can be explained quite simply:
The first step of any ML process is Data Collection. The quantity and quality of the data collected will dictate how accurate your model is.
The next step is Data Preparation, where the data is cleaned, randomized, visualized and wrangled (transforming and mapping data from one ‘raw’ data form into another format) in preparation for model training.
Choosing a Model is next, and will be determined by the task to be completed.
The fourth stage is Model Training, which is an iterative process which hopefully results in the model being able to answer a question or make a prediction correctly as often as possible.
Once trained, you need to Select the Model you intend to use by selecting the best model over the validation set.
Once the model is trained and the model is selected, you need to Evaluate the Model by using a metric or combination of metrics to measure the performance of the model in meeting its objective. This evaluation can consist of testing the model with data that it hasn’t seen before.
Parameter Tuning follows evaluation, in which the model’s parameters are tuned for improved performance. These parameters may include the number of training steps and the learning rate.
The final stage is Prediction Making, in which test data that has been withheld up to this point is used to test the model. This stage will give a better approximation of how the model will perform in the real world.
Okay, it’s all making sense now! What next for ML?
Despite all the buzz and hype that ML and AI has created already, it is clear that the field is at a relatively infant stage when you consider and appreciate the potential of the technology and how it can drastically disrupt the world, we live in. The technology will continue to develop and grow with better training and research into its capabilities and potential functionalities. Although we have been given various examples of how the technology can redefine various industries, it is evident that the full capabilities of the technology will not be felt for many years yet as experts and data scientists attempt to grasp a better understanding of the AI field.
There is no doubt that the future will be extremely interesting and exciting for those working in the AI field and for organisations who attempt to and successfully integrate the technology into their offerings. The various branches of AI will continue to grow, and new branches will emerge as times passes. Many organisations may feel they need to adopt AI and ML in order to remain competitive, and whilst this may be true in the long term, AI is better suited to solving complex issues using large buckets of data which obviously may not be feasible for some companies.
The future of AI and ML is unpredictable and dynamic, and who knows what direction the likes of Elon Musk might decide to take it. What is clear is that the algorithms that support for example, natural language processing (NLP) and speech recognition will only improve and as a result will become more integrated into our daily lives as technology users.
We hope this blog has provided you with a better understanding of what ML and AI is, whilst removing as much of the technical jargon as possible. As you can see, this blog only covers the tip of the ice berg, but the hope is that after reading this blog, your understanding of the topic and desire to further improve your knowledge of all things AI and ML has increased!!!