Mohammad Baker Ahmed Almousawi
19 Oct, 2020
What is Machine Learning?
Machine Learning is the subset of Artificial Intelligence that allows systems to have the ability to learn and improve without being directly programmed, enabling them to access past data and use it to continuously keep learning without a human intervening. It has been used to build mathematical models, making predictions from previous data and is now being used in areas such as email filtering, speech recognition and image recognition. Machine learning compromises of two main techniques; supervised and unsupervised learning.
- Supervised Machine Learning
This technique is when sample labelled data is provided to the machine to train it and consequently predict an output. Supervised Learning uses both input and output data to build a model.
- Unsupervised Machine Learning
Unsupervised Learning uses input data to gain insights from datasets without predetermined results. To be more specific, the data given has not been labelled or classified so the algorithm is required to work without any supervision.
Modelling or algorithms in machine learning is a simplification that attempts to demonstrate an aspect of the real world. An exact result is never produced but a prediction with a certain degree of confidence. How high the confidence is depends on how close a prediction matches with an existing example.
The history of Machine Learning
Machine learning began to become a reality around in the 1900’s, rather than being theories or concepts it was slowly turning into a reality.
From theories to reality
- In 1943 a neural network with electrical circuits was modelled by a mathematician and a neurophysiologist. This idea would then be applied by computer scientists in their work.
- In 1952 Arthur Samuel created a program that allowed an IBM computer to improve on checkers the more it played.
- In 1959 the first neural network was applied to remove echoes over phone lines.
- In 1985 an artificial neural network taught itself to pronounce 20,000 words in one week. This was invented by Terry Sejnowski and Charles Rosenberg.
- In 1997 IBM’s Deep Blue beat a chess master Garry Kasparov.
- In 1999 the CAD Prototype Intelligent Workstation reviewed 22,000 mammograms and detected cancer 52% more than radiologists.
Modern Machine Learning and its applications in the real world
- In 2006 Neural network research was rebranded as ‘deep learning’ by Geoffrey Hinton.
- In 2012 Google’s neural network learned to recognise humans and cats in YouTube videos. It had an accuracy of 74.8% in detecting cats and 81.7% in detecting faces.
- In 2014 a Chatbot managed to convince 33% of human judges it was a Ukrainian teen. Devised by Alan Turing in 1950, 60 years later it worked.
- In 2014, computers were able to improve ER experience in hospitals. Healthtech used event simulation to predict ER wait times, with these predictions hospitals were able to reduce wait times.
- In 2015, Google’s AlphaGo beat a professional player at Go. This is considered to be the hardest board game in the world.
- Also in 2015, Human detectives with the aid of a machine learning program geared up to take out criminals on the PayPal site.
- In 2016, North Face uses IBM Watson’s natural language processing in a mobile app. Consumers could find what they are looking for through a conversation with the digital personal shopper.
- In 2017 Alphabets Jigsaw team built a system that learned to identify trolling from reading millions of website comments.
The large technological advancements Machine Learning has led to, permitted it to become present all around us. Whether that is keeping us engaged on Netflix or aiding in the improvement of healthcare.
Machine Learning in Cyber Security
The incorporation of machine learning in cybersecurity has allowed them to prevent attacks and grant them the capability to respond to changes in behaviour through analysing patterns and learning from them. Cyber teams are now able to respond to active attacks and prevent possible threat, so time spent on routine tasks is significantly reduced and organisations can use their resources more strategically. The use of machine learning algorithms have been effective in identifying unusual behaviour and providing rapid action.
Machine Learning has the ability to sift through millions of files to identify the one that could be malicious and to uncover threats before it causes an enormous problem. There are a few areas machine learning has contributed assistance to, in order to handle cybersecurity threats such as spear-phishing, watering hole, webshell, ransomware and remote exploitation.
- Spear phishing
Traditional phishing detection techniques do not have the precision and speed to catch every malicious link, leaving many people at risk. Machine learning algorithms have led to predictive URL classification models that can find patterns to reveal what the email of the malicious sender was and with these models they can recognise behaviour such as designs or email headers to then identify whether an email is malicious or not. This solution prevents hackers targeting single individuals and then proceeding to scam them.
- Watering hole
Watering hole attacks are another example of targeted attacks machine learning can aid in eliminating. Watering hole attack is when an attacker tries to compromise a specific group of end users by infecting websites that the members are known to use, the end goal is to then infect the victims computer and gain access in the victims place of employment. ML algorithms can set and ensure the security standard of the web application services by analysing the path traversals of the website to detect whether users are led to malicious websites and detect malicious domains.
Ransomware is malicious software created to purposely block access to a computer until a sum of money is paid in exchange for the encryption key required to unlock the blocked computer and files. Data sets that are trained to analyse behaviours of ransomware attacks can detect unknown ransomware using deep learning algorithms. The algorithm would be required to find key features for every file in that data set which will then be used to train the model for the acquired data set. If in the future there’s a ransomware attack on the system, it will be checked with the trained model and that will allow the necessary actions to be taken before the whole system is blocked.
- Remote exploitation
Remote exploitation is a threat that targets an individual computer or its entire network through any vulnerabilities in order to gain access to the entire system. Attackers will use this technique of attack to steal or exploit sensitive data and cause severe damage by installing malicious software. They can do this in many ways such as denial of services attack, DNS poisoning or port scanning. The use of machine learning and trained algorithms will enable the analysis of system behaviour to identify unusual network behaviour to track down a possible exploitation before it occurs.
A webshell is a piece of code that can be uploaded to a webserver that enables remote access, allowing an attacker to gain full access to the database. Attackers often target the backend of online businesses where they will find and collect payment or credit card information of the customers. Machine learning models can be trained to identify malicious files and identify web shells to eradicate them before the system is breached.
The future seems vast for cybersecurity. With the help of AI and ML, experts will have the opportunity to analyse high volumes of data that can directly lead them to have accurate predictive analysis to foresee the next attack, reduce wasteful expenditures as developing techniques to secure data will not burden resources and malware or other infections will be easily detected because potential threats will be easily recognisable.
Almost all aspects of technology has weaknesses and strengths, and that includes using machine learning algorithms to combat cybersecurity threats. Introducing algorithms requires a human being to train the model system giving freedom for human error to occur and that may significantly reduce the overall accuracy in predictive modelling. Algorithmic bias is also a problem that needs to be tackled, if a machine is constantly fed one particular type of information then the decisions it makes will have a degree of bias without knowing the consequences of it – as it only knows what its taught. Making your system completely reliant on machine learning models gives attackers leeway to exploit the system by disrupting the models decision making or reverse engineering the model to near accuracy.
Is Machine Learning ethical?
Although the main ethical concerns are towards AI because of a potential job loss due to robots, the possibility of AI making mistakes with dire consequences and the ability for people to use them in malicious ways – machine learning also raises a few concerns.
Machine learning has had a significant impact on our society whether that is good or bad. It is important to understand that ML has the potential to be biased in many ways. Pre-existing biases can be incorporated into the data to train the algorithm which may cause sampling bias as a result of that. Prejudice bias can also become an issue due to the human input and whichever cultural stereotypes they will have can majorly skew the results and affect the predictive modelling. In 2015 Amazon had to scrap their AI and ML recruiting tool as it did not like women. Their computer models were trained to hire applicants based on observing patterns in the applications that were submitted over a 10 year period, this became an issue because the tech industry is male dominated and led the system to favour male applicants over the female ones.
Although bias in every instance poses as an issue, every issue can be solved. Dealing with bias when it concerns machine learning algorithms can be to use preventative measures such as training datasets to be more diverse, having a wider range of diversity in the AI/ML field or choosing the right learning model for the problem.
How are Tech companies using Machine Learning?
Tech companies acquiring lots of users, massive amounts of traffic and creating a large digital presence on the internet enables them to have an extensive access to data, providing them with the opportunity make good use of machine learning. So, how exactly are they using it?
Pinterest have invested in this technology to make their platform more efficient for users and to increase user engagement. They use ML to detect content that is spam, to develop ad performance and use predictive modelling to recommend suggestions to their users that they may be interested in.
Baidu is a Chinese search engine that is utilising ML to create innovative developments. Their newest development is ‘Deep Voice’, a neural network capable of generating human voices that are tricky to differentiate from real human voices. Deep Voice will lead to many beneficial uses such as real-time translation and voice search applications.
Google is considered to be one the most advanced tech giants in ML and AI. They use machine learning and predictive analysis to provide search results that are monetised through advertising. Google makes use of AI in their products such as Google Home and Google Assistant.
NASA implements machine learning to aid them in their further discovery of space. Their spacecrafts generate massive amounts of data which they use to identify patterns, monitor astronaut health in whilst in space and identify other planets that were previously undiscovered.
Netflix’s ability to recommend and suggest users TV shows or movies to watch is because of machine learning models that make use of user data such as watch history, so consumers can be kept engaged and continue with their subscription. The creation of the ‘personalisation’ feature has also significantly improved streaming quality.
The implementation of machine learning on the twitter platform enables them to drive engagement, showcase content that is relevant to their users and combat hate speech that is prevalent on their website and app.
What does the future for Machine Learning look like?
The growth of machine learning has been astronomical with its market at a compound annual growth rate of 44.1% and expected to grow up to 8.81 billion by 2022. This rapid growth is largely due to the proliferation in data generation and technological advancement. The future seems vast and a few areas will be able to benefit from these progressions.
- Environmental solutions
Environmental issues affect people globally whether it be forest fires, tornados or thunderstorms. With ML and the masses of data that is stored, trends can be identified to provide accurate solutions and plan ‘what if’ scenarios to avoid massive environmental tribulations.
- The use of Quantum Computing
The increased adoption of quantum computing will allow faster data processing when used alongside with machine learning, this will aid in enhancing analysis and pulling insights from datasets. The increase in performance will allow businesses and organisations to accomplish major results compared to if they had just used machine learning methods without incorporating quantum computing.
- The entertainment industry
Machine learning has already found its way in the entertainment industry with streaming services such as Netflix, Amazon Prime and Google play relying on algorithms to give consumers the best quality possible. Endless data about peoples viewing habits and what they like to watch are used to offer recommendations on what else they might find enjoyable and entertaining – with this information companies are able to provide enhanced personalisation to their target audiences which ultimately keeps us all coming back.
- Improved healthcare
The healthcare industry has and will continue to utilise machine learning to help treat their patients that have hospital related illnesses. The use of deep learning models and predictive modelling will make it possible for doctors to diagnose patients faster, reduce the costs of healthcare and develop new drugs.
- Security in banks
Fraudulent behaviour has been a constant issue for banks and the customers that become victim to these behaviours. Using location data and purchase patterns, machine learning based anomaly detection models will be able to track anomalous behaviour and inform you instantly of the unusual activity, meaning you could be saved from potential theft.
Jenny is a Security Web Intelligence platform created to bring users and cyber security vendors together. She uses AI and Supervised Machine Learning to search and find security vendors from the internet, to then index them into categories. Because of Jenny you can access 1000’s of those companies granting you entry to key information you may require about the company you have in mind. With Jenny’s help, you can find the right vendor for you, ensuring you get the cyber security products and services essential for your individual needs.
- A history of machine learning
- A Short History of Machine Learning — Every Manager Should Read
- Machine Learning for Cybersecurity 101
- What is Machine Learning? A definition
- What Is Machine Learning? 3 things you need to know
- A BRIEF HISTORY OF MACHINE LEARNING AND DATA SCIENCE
- AI and machine learning in cybersecurity: Trends to watch
- Machine Learning in Cybersecurity – Hype vs. Reality
- Machine Learning for Cybersecurity: Good, but Imperfect
- How will AI and Machine Learning Affect Cyber Security?
- Machine learning: Is it ethical?
- Machine learning ethics: what you need to know and what you can do
- What are some ethical issues regarding machine learning?
- What is the Future of Machine Learning?
- 15 Ways Machine Learning Will Impact Your Everyday Life