Machine Learning: Evolution, Objectives, and Applications in Material Discovery - Research Paper Sample

1: Introduction

1.1 History of Machine Learning

By its nature, artificial intelligence is a sub-field of machine learning and computer science development in facilitating the systems in learning computational theory and pattern recognition. Rather than following static programs, machine learning incorporates machines to learn and construct algorithms to make predictions from the available data sets without being explicitly programmed. However, the principle behind machine learning has a long history, relying on mathematics and statistics from many years ago, along with the enormous changes in computing power.

Is your time best spent reading someone else’s essay? Get a 100% original essay FROM A CERTIFIED WRITER!

Save 25%

In the early 1940s, Walter Pitts and Warren McCulloch, a neurophysiologist, modeled the neural network based on an electrical circuit. In 1950, the Turing Test was invented by Alan Turing to differentiate between a human and a computer. In subsequent years such as 1958, Frank Rosenblatt designed the first artificial neural network, known as Perceptron, a simplified mathematical model of the brain's neurons that could learn to recognize patterns and shapes. Architects built this model on the work of Walter Pitts and Warren McCulloch, also known as, McCulloch-Pitts but it lacked a mechanism for learning, which is crucial for it to be usable for AI.

In the following year, 1959, Bernard Widrow and Marcian Hoff created two models with the name of ADELINE and MADELINE, which could detect binary forms and eradicate the echo on the phone line; respectively, both are very useful and in use today. There is not much that happened during the 1980s and 1990s, which could be significant. However, in 1997 the public awareness of artificial intelligence increased considerably when the IBM computer Deep Blue beat the world chess champion, Garry Kasparov, in the match, yet the program relied on brute computing power to determine its path checkmate.

Geoffrey Hinton's introduction of the term 'deep learning' rebranded it as the neural networks' name in 2006. He is known as the godfather of deep learning as he is the pioneer in researching the neural networks that now underlie much of artificial intelligence. Throughout the 21st century, face recognition has become a reality via backpropagation. Deep learning neural networks and machine learning could be improved vastly for many tasks, such as shape recognition, word prediction, and many more.

Graphic processing units (GPUs) and Convolutional Neural Networks (CNNs) were being widely used in machine learning when AlexNet won the ImageNet competition in 2012, where they also created ReLU that momentously improves the efficiency of CNNs. ImageNET is the subset of a public computer vision dataset with an extensive collection of annotated photographs intended to develop the computer vision algorithms, foster the development, and set the state-of-the-art algorithms' benchmark.

Facebook then developed software that can recognize and verify individuals on photos using the DeepFace algorithm. Another well-known example is Facebook's News Feed, where the software uses statistical and predictive analysis to recognize the usual trends in the user's data and then create the News Feed. This program will adjust accordingly based on the user's activity.

In 2014, Google acquired DeepMind Technologies with DeepMind programs that developed the neural network that learned to play video games by analyzing the behavior of pixels on the screen and accessing the external memory known as the Neural Turing Machine. OpenAI was initiated by Elon Musk and others in 2015 to create safe artificial intelligence that can benefit humanity. Microsoft also, in the same year, created the Distributed Machine Learning Toolkit, which enables the efficient distribution of machine learning problems across multiple computers at once

1.2 Objectives

This project aims to tap into the application of inverse molecular design using machine learning for matter engineering in material discovery. The following are the objectives of the work:

1. To study the concept of machine learning in inverse design and be applied in the project.
2. To perform a different number of iterations and differentiate the different outcomes.
3. To examine and analyze the model accuracy and model loss of the samples.

1.3 Thesis Outline

There are five (5) chapters in this research that are summarized and analyzed closely. Each chapter is briefly justified as follows.

Chapter one (1) consists of the background, history, and introduction of machine learning. The paper then identifies three objectives in which the primary goals account for their existence. Then, the paper elaborates on the scope of the study briefly.

Next, Chapter Two (2) provides an outline, portrayal, and past research assessment related to inverse design using machine learning. This research paper utilizes materials that respect the procedure and methodology of this study and its topics.

In Chapter Three (3), the paper fonts and explains an extended clarification of the ways, techniques, and instrumentation used in this study. The methods will be explained thoroughly through the step-by-step method, from setting up the computer, training the program, and expected results for the objectives stated.

Chapter four (4) comprises the results and analysis. The researcher then compares and tests the outcomes from the program run. Thus, the paper explains every finding in its entirety.

Lastly, Chapter Five (5) points out the conclusion, a summary of this study's overall outcomes. It then presents the last judgment with a recommendation for future work.

CHAPTER 2: Literature Review

Tom Mitchell from Carnegie Mellon University said "A computer program is known to learn from experience (E) along with certain task (T) and some performance (P). Each performance (P) for each task (T) improves with the number of experiences (E)".

2.1 Machine Learning

Machine Learning is the computer systems' ability, where designers apply algorithms to learn from data and information autonomously. It is not explicitly needed for the computer to be a machine learning program as the algorithms can be modified and improved on their own (Marr,2016). Machine learning allows computers to execute the desired tasks intelligently by learning independently from the data (Hosch, 2016) rather than following pre-programmed instructions (The Royal Society, 2017).

Machine learning is one of the most exciting recent technologies in Artificial Intelligence present in many applications daily. Every time individuals use a web search engine like Google or Bing to search the internet, one of the reasons for its functionality is that a learning algorithm, one implemented by Google or Microsoft, has learned how to rank web pages. Additionally, the use of Facebook to recognize a friend's photo is an act of machine learning. Spam filters, another example of a machine learning algorithm in emails, save the user from wading through tons of spam emails. In this paper, a brief review and prospect of machine learning's vast applications have been made.

According to Arthur Samuel, Machine learning is defined as the field of study that gives computers the ability to learn without being explicitly programmed. Arthur Samuel was famous for his checkers playing program. Initially, when he developed the checkers-playing program, Arthur was better than the program. Nevertheless, over time, the checkers playing program learned the right board positions and terrible board positions by playing many games against itself.

Tom Mitchell gave a more formal definition of machine learning as the performance (P) of the task (T) improves with experience (E). In the checkers playing example, the experience (E) was having the program play games against itself. Task (T) was playing checkers where the performance measure (P) was with the probability that it won the next game of checkers against some new opponent. There are more extensive and more massive data sets that designers can now understand using learning algorithms in all engineering fields.

2.2 Machine Learning Applications

Machine Learning has revolutionized the world hence transforming the approach in scientific research, catalyzing the fourth epitome shift of scientific discovery through the emergence of data-intensive science over the past two decades (Hill et al., 2016). In overcoming the technical issues in this modern industry, the finding of high-performance functional materials is vital. There is good reason to believe that intelligent data analysis will become inescapable as an essential ingredient for technological advancement. Machine learning enables man to communicate with people, publish news and sports matches, or even self-driving cars using computers.

There are three categories of machine learning applicable today: supervised, unsupervised, and reinforcement machine learning. Supervised ML utilizes labeled data to train machines, with the aid of an already provided small dataset. On the other hand, unsupervised data holds the advantage of being able to utilize unlabeled data; thus, no human labor is required to generate the machine-readable dataset, thus providing more room for extensive dataset...