This report is authorised by Oscar Imaz Mairal, Course Coordinator of BUS 108 Introduction to Informatics, at the University of the Sunshine Coast, Queensland, Australia. This report has been compiled to meet the requirements of the report assignment for this course.
This report is limited to researching and formatting skills in the first-year course BUS108 Introduction to Informatics at the University of the Sunshine Coast.
This report looks at the importance of artificial intelligence as machine learning and the major types of machine learning. Specifically, deep learning is investigated in depth to discover two possible recurrent neural network business models. Academic sources including the course textbook, two provided papers and additional peer reviewed and non-peer reviewed journal articles are used to provide theory and evidence. Innovative recommendations are provided regarding how to improve the two proposed possible recurrent neural network business models.
To complete this report secondary data from reputable academic sources, including the course textbook, two provided papers and additional peer reviewed and non-peer reviewed journal articles have been researched and analysed.
2.0 Artificial intelligence as machine learning
2.1 Definition of artificial intelligence as machine learning
Artificial intelligence (AI) can be defined as the capability of a machine to execute cognitive functions associated with human minds (Chui, et al., 2018) (Grupe, et al., 1995). Such includes perceiving, reasoning, learning, and problem solving (Grupe, et al., 1995) (Neapolitan, 2018). Some instances of technologies that allow AI to rectify business problems are robotics and autonomous vehicles, computer vision, language, virtual agents, and machine learning (Grupe, et al., 1995) (Reynolds & Stair, 2018).
Recently developments in AI have been accomplished by applying machine learning to data sets that are substantial in size (Grupe, et al., 1995) (Reynolds & Day, 2018). Machine learning algorithms process data and experiences by distinguishing patterns and learning how to make predictions and recommendations (Ghahramani, 2015) (Grupe, et al., 1995). This is in contrast to just simply obtaining unambiguous programming instructions. The algorithms used also adapt in response to new data and practices to improve both efficiency and effectiveness over time (Grupe, et al., 1995) (Reynolds & Stair, 2018). Machine learning provides both predictions and prescriptions, with the types of analytics detailed in figure one (Grupe, et al., 1995).
2.2 Description of the major types of machine learning
There are three major types of machine learning, those being supervised learning, unsupervised learning and reinforcement learning (Brynjolfsson & Mitchell, 2017).
Supervised learning is an algorithm that adopts human training data and feedback to discover the relationship between given inputs and outputs (e.g. how inputs such as the time of year and current interest rates predict outputs such as housing prices) (Alpaydin, 2014) (Grupe, et al., 1995). Supervised learning is used when how to classify the input data and the type of behaviour wanted to predict is known, but an algorithm is needed to calculate such on new data (Chui, et al., 2018) (Grupe, et al., 1995). Supervised learning works by a human labelling every element of the input data (e.g. in the case of predicting house prices, input data is labelled as time of year and interest rates) and defining the output variable (e.g. housing prices) (Grupe, et al., 1995) (Langley, 1986). The algorithm is then trained on the data to find the connection between input and output variables (Grupe, et al., 1995) (Neapolitan, 2018). Once the algorithm is sufficiently accurate training is complete and the algorithm is applied to new data (Ghahramani, 2015) (Grupe, et al., 1995).
Unsupervised learning is an algorithm that investigates input data in the absence of a specified output variable (e.g. the exploration of customer demographic data to identify patterns) (Chui, et al., 2018) (Grupe, et al., 1995). Unsupervised learning is used when it is unknown how to classify the data and the algorithm is wanted to find patterns and classify the data. Unsupervised learning works by the algorithm receiving unlabelled data (e.g. a set of data describing customer journeys on a website) (Grupe, et al., 1995) (Langley, 1986). A structure is then inferred from the data and the algorithm is used to identify groups of data that exhibit similar behaviours (e.g. customers that demonstrate similar buying behaviours) (Reynolds & Day, 2018) (Grupe, et al., 1995).
Reinforcement learning is an algorithm that studies how to execute a task by simply aiming to maximise rewards received for its actions (e.g. maximising points it receives for increasing returns of an investment portfolio) (Alpaydin, 2014) (Grupe, et al., 1995). Reinforcement learning is used when lacking in training data, the ideal end state cannot be clearly defined or the only way to become educated about an environment is to interact with it (Brynjolfsson & Mitchell, 2017) (Grupe, et al., 1995). This algorithm makes an action on the environment in question (e.g. makes a trade in a financial portfolio) and will then receive a reward if the action acts as a catalyst to bringing the machine closer to maximising total rewards available (e.g. the highest total return on the portfolio) (Ghahramani, 2015) (Grupe, et al., 1995). The algorithm will correct itself overtime to optimise for the greatest series of actions (Grupe, et al., 1995) (Neapolitan, 2018).
3.0 Deep learning
3.1 Definition of deep learning
Deep learning is a subfield of machine learning that is able to manage a wider range of data resources, necessitates less data pre-processing by humans and has the ability to provide more precise results than traditional machine learning approaches (Grupe, et al., 1995) (Porter, 2016). Within deep learning, interconnected layers of software-based calculators known as neurons form a neural network. This allows the network to consume large quantities of input data and process them through multiple layers that allows increasingly intricate features of the data to be learnt (Grupe, et al., 1995) (Langley, 1986). As a result, the network is then able to make a determination about the data, learn if its determination is correct, and use what it has been learnt to make determinations about new data (e.g. once the network ascertains what an object looks like, it can recognize the object in a new image) (Lecun, et al., 2015). Deep learning has the potential to outperform traditional methods such as image classification, facial recognition and voice recognition (Grupe, et al., 1995). There are two major deep learning models, convolutional neural network and recurrent neural network (Patterson & Gibson, 2017).
3.1.1 Recurrent neural networks
A recurrent neural network is a multilayered neural network that is able to store information in context nodes (Grupe, et al., 1995) (Keller, et al., 2016). This allows the network to learn data sequences and output a number or another sequence (Bitzer & Kiebel, 2012) (Grupe, et al., 1995).