Machine Learning-An overview.
Let’s explore the Fundamentals of ML.
The purpose of this article is to educate you about different Machine Learning systems, their uses and any challenges that need to be overcome.
What is Machine Learning?
Machine learning is the study of computer algorithms that improve automatically through experience and by the use of data.
Source:Wikipedia.
Written simply,ML is the science-and art-of programming a computer to learn and self improve from data sets that it is fed(called the training set) , in order to provide a solution to a task.
If you were to download an Encyclopedia onto your device,it would store an abundance of data,however, it isn’t helping to fix a problem.Therefore,this would not be an example of Machine Learning.
What’s the Difference between AI,ML and DL?
Machine Learning is a subset of Artificial Intelligence and Deep Learning is a subset of ML. Deep Learning is used to mimic how the human brain works- in order to process data,create patterns and for use in decision making. Deep learning is being used most commonly today,in virtual assistants like:Alexa, Siri and Google Assistant to create almost human-like interactions between the device and the customer.
What are the different types of Machine Learning Systems?
There are perhaps 14 different types of Machine Learning Systems but for the purposes of this article,these can be broadly categorised based on :
❇️Whether or not the algorithm is being trained under human supervision(Supervised,Unsupervised,Semi-supervised and Reinforcement Learning).For example, a spam filter is trained with many emails which will have been labelled as ham(normal) and spam. These classes will help it learn to classify new emails as well as learn from each training instance(each email it is fed).This is an example of Supervised Learning.
❇️Whether or not the algorithm can learn in small increments,very quickly(Online vs Batch learning). Online Learning is useful for systems that receive data as a continuous flow-e.g. stock prices.
❇️Whether or not the algorithm compares data points, or instead, detects patterns within the training data to build predictions(Instance based vs Model-based Learning).
Why is Machine Learning becoming increasingly more relevant?
As the digital divide gradually decreases, the amount of data being stored in data centres across the planet is ever increasing. We need ways of managing this data automatically and this can be achieved through Machine Learning.
Another reason ML is becoming more relevant, is because modern computers are significantly more powerful than the very first machines. Even the latest Iphone has more computing power than the Apollo Guidance Computer,which was used to provide computation and electronic interfaces for guidance, navigation, and control of the spacecraft! ML algorithms have also progressed over time-they are continually being improved,as they learn from each training instance.
Examples of ML in day-to-day life.
It’s likely that you have already used an ML algorithm,without even realising! Machine Learning algorithms are being used to:recommend purchases online;power self-driving cars;detect fraud;recommending you accounts to follow on social media platforms like Instagram and in many other ways!Streaming services like Netflix and Amazon Prime Video already use an ML algorithm to provide you with film/series recommendations. Here’s how they work:
Once a customer finishes watching a film/series they have the option to give it a ‘thumbs up’. This action feeds data to the algorithm,informing it that this particular film/series is in accordance to their preference. It will then use this data to calculate how similar their preference is,to other users on the database. For example, if you ‘liked’ Shadow and Bone,then your preference will be considered closer to other users who also liked the series but further away from users who gave it a ‘thumbs down’ or a ‘dislike’. The algorithm will then suggest you shows based on these preferences. These preferences are quantified by the performance measure. The name of this algorithm is called nearest neighbour, because it uses information about your nearest neighbours to give you shows that will match your taste! This particular example uses Supervised Learning-as mentioned earlier.
What challenges need to be overcome?
In 2015, an image recognition trial was ran to pick out various objects out of a group of pictures. The algorithm proved to be 95% effective at identifying objects-compared to 94%, identified by humans. However, understanding context,like what each object is doing, can still be very challenging.
Below is an example of a Deep Neural Network system being used to identify animals:
As you can see, after adding some ‘noise’ (slight changes in the image,that are almost invisible to us) the algorithm was fooled to identify the images incorrectly. If the noise had been added to the algorithm ran in 2015, the humans would have definitely identified the objects correctly!
ML also has various ethical implications; an example of this was when a study by Stanford University resulted in an ML algorithm which could correctly identify the sexual orientation of gay and straight men 81% of the time, and 74% for women. The training set involved more than 35,000 facial images that men and women publicly posted on a US dating website.Human judges performed much worse than the algorithm, accurately identifying orientation only 61% of the time for men and 54% for women. The ethical implications of ML like this is that, it has the potential to violate people’s privacy or be abused for anti-LGBT purposes.
Therefore, it is vital that Machine Learning is being used to properly benefit society in healthcare,transport and many other industries!
Written By: Muskan Sharma
References:
Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems
[Book by Aurelien Geron]
