What is machine learning?
According to the Oxford Dictionary, machine learning is the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw interferences from patterns in data. Machine learning is a subfield of artificial intelligence in which the goal is to understand structures of data and fit that data into models that are able to be used by people. Any and every technology user has benefitted from machine learning, such as facial recognition or even recommendation engines that suggest a show or movie to watch next. Machine learning has been around for probably much longer than you'd guess, 1943 to be exact. Machine learning was first conceived from the mathematical modeling of neural networks, so its been around for quite some time, but recently has become a bit more relevant with storage becoming significantly cheaper.
Machine learning is great for solving many types of complex problems that would be too much for a human to calculate on their own. Some other relevant examples of machine learning are:
- Medical Diagnosis:
Machine learning can help us with diagnosing diseases in patients. Some ways machines assist us are in formulating diagnosis as well as recommendation of a treatment option. Machine learning can also assist in the analytics of bodily fluids, and in the case of rare disease, facial recognition software working in tandem with machine learning can be used to scan patient photos and identify phenotypes that correlate with rare genetic disorders.
- Predictive Analysis
Predictive Analysis in machine learning can help determine whether transactions are fraudulent or legitimate as well as improving prediction systems abilities to calculate the possibility of a fault.
- Speech/Image Recognition
Speech and Image Recognition coupled with machine learning is probably the most familiar of these. Speech recognition can be used to translate speech to text or text files, help with appliance control, and facilitate voice searches. Image recognition is equally prevalent in facial recognition in cell phone locks, assigning names to images(like tagging a friend on social media), and analyzing handwriting.
Machine learning is segmented into several different techniques, mostly divided into 4 categories. These are supervised learning, unsupervised learning, reinforcement learning, and semi-supervised learning. Supervised learning is applicable when a machine is given sample data as well as output data with correct labels. This allows the machines to basically "study" correct images while making predictions about output values and errors that could occur, and attempting to correct them through algorithms. Unsupervised learning is when a machine is trained with some input data, but the output is not known. This is not used as commonly in practical business, but it does help in exploring the data and can infer from the data it is given to describe hidden structures from unlabeled data. Reinforcement learning is feedback-based machine learning, where computer programs need to explore and observe their environment. For each good or correct action, the machine will get a positive reward, and for each incorrect response, they will get a negative reward. The goal of the program is to maximize the positive rewards, and since there is no labeled data, the machine has to rely on learning through experience only. Lastly, semi-supervised learning is an amalgamation of both supervised and unsupervised learning. It performs actions on a mixed set of data, some being labeled and some being unlabeled. This helps to reduce the costs of the machine learning model(not having to label everything), and also helps to increase the accuracy of the machine learning model.
Machine Learning's Most Interesting Paradox
Moravec's paradox is a phenomenon in machine learning which occurs when we use AI-powered tools. It states that tasks which humans find simple due to sensorimotor skills that humans naturally possess are difficult things for machines to learn. Additionally, the inverse is true, tasks which are complex to humans, such as solving complex logic puzzles and performing advanced mathematics almost instantaneously, can be relatively easy to teach to a machine. Machines have trouble learning seemingly "simple" tasks because the things that we take for granted, or deem obvious, came about from thousands of years of evolution. At the bottom of the article, I have attached the YouTube video I watched that made me interested in learning more about machine learning and Moravec's Paradox. If you have a few minutes to spare, I would highly recommend watching it or bookmarking it for later.
Challenges in Machine Learning
Some of the most common challenges when it comes to machine learning are lack of quality data, the ability to understand which processes need automation, implementation, lack of skilled resources, and inadequate infrastructure. Some of these challenges are due to machine learning as a whole requiring vast amounts of data churning capabilities, as well as machine learning itself to still be in early stages of development. This means not only is there a shortage of infrastructure and data, but also that there is a shortage of skilled employees available to manage and develop machine learning and analytical content. One of the biggest challenges machine learning faces is bad data. Any sort of incomplete or irrelevant data is malicious to machine learning, so professionals must take time to rigorously evaluate the data before it is used in machine learning.
Hopefully you've learned a thing or two about the basics of machine learning as it is quite the fascinating field, and I'm looking forward to diving deeper into the basics and learning more about machine learning as it develops. Hope you enjoyed the read!
Video:
Moravec's Paradox Explained in 5 Levels of Difficulty
Sources:
Top comments (0)