The hype around AI continues. Deep learning has become almost a synonym for AI despite the so many other techniques that exist within the machine learning field.
So, if you are trying to apply deep learning, would not you want to know if deep learning is appropriate for your application? In this article you will learn about deep learning and the three simple questions to ask before you start.
What is deep learning?
Deep learning is a subset of machine learning. It is based on artificial neural networks that mimic biological neural networks of the human brain. At the basic level is the perceptron and the mathematical representation of a biological neuron. Like in the biological human brain, several layers of interconnected perceptrons could be found in the artificial neural networks.
Input values get passed through the hidden layers of the network until they converge to the output layer. The output layer represents the prediction outcome. Each node of the hidden layers has a weight, and it multiplies its input value by that weight. There are several types of artificial neural networks such as convolutional neural networks and recurrent neural networks.
Deep Learning has significantly improved the accuracy of the machine learning models and has made great advances in fields such as computer vision and speech recognition.
What are deep learning applications?
Deep learning has been applied to several areas including:
- Healthcare: deep learning is having a significant impact on healthcare, applications range from cancer detection, digital consultation to precision medicine. There several success stories of deep learning cancer detection accuracy:
- Skin cancer melanomas: machine find 95% of melanomas vs. 86% dermatologists
- Breast cancer: researchers at Imperial College London are working with DeepMind Health to develop AI-based techniques to improve the accuracy of breast cancer screening.
- Colorectal cancer: detects early stage with 86% accuracy.
- Autonomous cars: deep learning is applied to train autonomous cars to drive by using large set of data from the digital sensor systems. The most popular approach is the rule-based approach where deep learning is applied to recognise road conditions including cars and passengers and handover the information to the processing unit to decide based on a set of pre-programmed rules.
Other approaches include letting the deep learning model to build its own decision-making capabilities.
- Machine translation:
Machine translation is the use of computers to automatically translate a given word or a sentence from one language to another. Deep learning has made significant contribution to machine translation. Deep learning has been applied to improve the statistical machine learning translation components and map source to target languages.
What are deep learning pros and cons?
Deep learning has made significant contribution to machine learning field and has dramatically improved accuracy rates. This is particularly relevant in image and voice recognition. Another advantage of deep learning is that it reduces the need for features engineering.
However, deep learning has limitation including the need for large amount of data. It also requires significant number of computations but with the advance of computer power that we have now a day this is not seen as a big limitation. Perhaps, the biggest limitation that deep learning has is the lack of transparency and the inability to fully understand how the outcomes where derived. This could be an issue when deep learning is applied to heavily regulated sector or where the application has ethical considerations.
So, when is deep learning appropriate?
To ensure that deep learning is appropriate for your application, you need to ask these three simple questions before you start:
- Question 1: Is the problem complex? The higher the complexity of the problem the more likely that deep learning could help.
- Question 2: Is there a large set of data? The smaller the data set the more likely that deep learning is not suitable.
- Question 3: Is there a need to explain relationship between variables? If there are stringent requirements to explain how outcomes were derived and the relationships between variables, then deep learning on its own is not suitable.