Over the last forty years, there has been a great advancement in digital information system hardware, and information technology. This led to the emergence of a fourth industrial revolution that continued the work done on the third revolution (Badr, M., Abrian, A. 2020). The recent breakthroughs have been faster than ever before. Thus, impacting the way many sectors implement their processes to adapt to their fast-changing environment.
One of the impacted sectors is the finance sector. Technological advances have made it possible to restructure the sector and the established players in it by creating new products and opportunities.
Fintech companies that emerged in the late 80s are nowadays one of the key players in finance. They have spread increasingly after the 2007 financial crisis.
The changes in technology have made it possible for fintech to penetrate the customer-focused areas. Thus, improving the quality of service, increasing profanity and optimising efficiency.
However, despite the potential of this transformation, there are a couple of drawbacks. One issue is the increased risk level of cybersecurity threats. As financial institutions implement more digital technologies, they become more vulnerable to data breaches, hacking, and all other types of cyber-attacks (Gunjan, S., Satish, K. 2020).
Moreover, the speed of the changes has made it harder for some employees to continuously adapt to using new systems and technologies. This has caused many companies to struggle getting rid of outdated legacy systems.
Ensuring proper regulations and legislation is crucial to helping finance companies to effectively and efficiently adapt to the fourth industrial revolution.
Schwab, K. (2016) The Fourth Industrial Revolution: what it means, how to respond. Available from: https://www.weforum.org/agenda/2016/01/the-fourth-industrial-revolution-what-it-means-and-how-to-respond/ [Accessed 23 October 2024].
Badr, M., Abrian, A. (2020) Industry 4.0 and its Implications for the Financial Sector. Available from: https://www.sciencedirect.com/science/article/pii/S1877050920323371 [Accessed 23 October 2024].
Gunjan, S., Satish, K. (2020) A decision-making framework for Industry 4.0 technology implementation: The case of FinTech and sustainable supply chain finance for SMEs. Available from: https://www.sciencedirect.com/science/article/abs/pii/S004016252200213 [Accessed 23 October 2024].
The last few years have witnessed a great evolution in technology and a rise of artificial intelligence. Models such as ChatGPT or Google Gemini have demonstrated a revolutionary capability in interpreting and processing human language, thus changing the way various industries condone their processes.
AI writers such as the one mentioned above have great benefits that will play a key role in reshaping the future. According to Hutson, M. (2021), the ability of these models to accurately summarise large amount of content will help speeding up the academic research process by saving the writers time to focus on reading all the material. Additionally, these models are trained to understand multiple languages which will allow the writers to reach a wider audience that is not native to their mother tongues. Another advantage is for the software engineers, as these models have proven to be valuable in helping to debug code and identify the issues within it, thereby improving the team's productivity.
However, a couple of drawbacks can be mentioned. Katy, I. (2023) discussed how copyright issues will be a future dilemma because the generated material cannot be fully attributed to the writers (Katy, I. 2023). Another noticeable issue is the harmful content that can be produced including hate speech, misinformation, or even dangerous commands. Ethical considerations are also being discussed such as the decreased creativity of humans in the long term.
In the long run, the usage of such models can be beneficial, but only if proper regulations are put in place. Ensuring transparency and the use of non-biased data will help build trust and help control any negative outcomes.
Hutson, M. (2021) Robo-writers: The rise and risks of language-generating AI. Nature, 591(7848), pp.22-25. DOI: https://www.nature.com/articles/d41586-021-00530-0 [Accessed 5 January 2025].
Katy, I. (2023) AI and the Writer: How Language Models Support Creative Writers, 2(3), pp.1-24. DOI: https://www.nature.com/articles/d41586-021-00530-0 [Accessed 6 January 2025].
Flanagin, A., & Kendall, T. (2023). Guidance for Authors, Peer Reviewers, and Editors on Use of AI, Language Models, and Chatbots. JAMA 330(8). DOI: https://www.sciencedirect.com/science/article/pii/S2590005622000911 [Accessed 6 January 2025].
For this activity we explored a tutorial on data analysis using Google collab. The following image showcase the data distribution of the AutoMpg dataset.
This activity consisted of running multiple Google colab programs to understand the different challenges possible in the raw data and the way they are handled. The figure below is a result of building a linear regression model using the Pearson's correlation measure.
We were asked here to calculate the Jaccard coefficient for a couple of pairs in the table below:
The activity was to observe the process of the algorithm to classify the data into different clusters. The figure below showcases the clustering animation with the centroid initialisations starting from the 4 top-most points.
A couple of activities were run in python to explore the different neural architectures with varied numbers of layers. The figure below shows the evolution of the error metric after each iteration (epoch) in a multi-layer perceptron.
For this activity, we ran a tutorial for the gradient cost function while doing some hyperparameter optimisation to observe how the cost decreases:
Original 100 iterations/ learning rate 0.08
50 iterations/learning rate 0.08
200 iterations/ learning rate 0.16
Doubling the learning rate and iterations does not always improve gradient descent, just as halving them might prevent it from reaching the minimum. The most effective approach is to monitor convergence and increase the number of iterations only if the process has not fully converged.
A convolutional neural network model was run using the CIFAR10 dataset and we got to explore the practical implementation of it in python using Google Collab. The model has a prediction accuracy of 63% on the test data with only 6 epochs. A couple of images were fed to the program to predict them, and we can see that it was able to correctly do so:
A couple of model performance measurement techniques were explored such as f1-score or r-squared. Additionally, a confusion matrix was built to show the true and false positives and negatives to see how a certain model can affect real-world applications.
Before the start of the module, I had limited experience using the recently known language models such as ChatGPT or Gemini. However, the process of how these models learn and can perform at such accuracy was a vague subject.
In the first unit of the module, the focus was on introducing machine learning, its past and future and the role it will play in reshaping human life. The lecture cast extinguished the difference between traditional software engineering and machine learning. It was interesting to learn about the automation solutions that the technology offers along with the technical aspects of the learning categories. The first chapter of the introduction to machine learning was a critical part of the reading material for this unit (Miroslav, K. 2021). It has helped me clearly understand the process of feeding learning models training data in form of labelled images and how they are classified based on their attributes.
The second unit Introduced the exploratory analysis which is a valuable part of the model-building process (Patil, S., Nagaraja, G. 2020). Although a previous module has covered it briefly, exploring it in detail clarified a few concepts. The fourth section of the first chapter in the introduction to machine learning discussed how it is important to visualise the data and understand the weaknesses within (Miroslav, K. 2021). It was interesting to learn about all the implications and effects they can have on the results of the model. The seminar explored the data analysis tutorial on Google Colab and we had a closer look at the process in a more technical way.
The third unit had insightful details about some of the most common statistical techniques to describe the relationship between the data variables which are correlation and regression. The lecture cast highlighted the mathematical approach to each technique and perfectly explained how to interpret the results in a simpler manner. The machine learning tutorial from CodeBasics showcased the methods and libraries available to use to replicate the process in coding, particularly in Python which has the easiest implementations.
The fourth unit introduced an open-source library called Scikit-learn. Although I was previously unfamiliar with this name, it was very enjoyable to visualise the process of performing linear regression on the given datasets in the activity. The reading material included the user guide of Scikit-learn and how it helps with supervised learning. It is helpful to see how varied the components of the system are.
The fifth unit delved into the basics behind clustering and the different techniques employed in evaluating the clusters and interpreting the results. The lecture cast clarified the logic and perfectly explained the evaluation process and the different clustering techniques. Chapter six of the introduction to machine learning briefly talked about the multilayer perceptron and what makes it convenient for the artificial neural network’s architecture (Miroslav, K. 2021).
The sixth unit had more details about the implementation of clustering using python. It was interesting to see how many libraries were available and the role they play in making this accessible as an open source. The group project was due by this unit and the pressure started. Although we had different time zones and personal life schedules, it was not hard to have an effective plan between the group members and the project was delivered on time. The experience was seamless, and I was happy to analyse the data for a business proposal in a group setting as well as my contributions in terms of business question and linear regression analysis.
The seventh unit was a continuation of the learning acquired in the previous reading material. There was an insightful explanation of the basics of human minds and how they work in the lecture cast. Additionally, the analogy with artificial neural networks was delved into among the different functions used such as activation functions. The python activities for this unit helped understand the effect different layers of a neural network model can affect the output accuracy.
The eight unit was quite enjoyable as it finally made it clear for me how the models are trained using deep learning. The lecture cast explained the backpropagation process and how it helps the model learn from its mistakes by adjusting the neurons’ connection weights after each iteration. The gradient cost function activity cleared some confusion about how to achieve a minimum error loss value while keeping the accuracy high.
The ninth unit was challenging compared to the previous one. The concept of convolutional neural networks was simpler to understand. However, the algorithm behind it was difficult to digest. The lecture cast came in handy in that way, it has simplified each step of the process from object detection to filter learning and filtering to pooling and ending with the output.
The tenth unit was even harder in the aspect of understanding CNN. Visualising learning cycles is always a fun and interactive method to gain in-depth understanding of a particular subject. The seminar content for this week was very helpful in getting ready for next week’s submission and equipped us with all the necessary information.
The eleventh unit allowed me to understand the different techniques used to select a specific model and evaluate its performance. The lecture cast was informative in citing popular methods such as cross-validation and the different classification metrics. However, the focus this week was on delivering the summative assessment. Being able to produce an image classification model that is trained and evaluated was a great experience in the end as it made me more comfortable building such systems and ready to work on real-world implementations.
The final unit concluded the module with an insightful visualisation of the future of machine learning and the industry 4.0 revolution. The reading material contained an interesting article about the new trends in machine learning and gave a clear idea about how the future would look like for the industry. The process of building the e-portfolio was simpler in this module after having most of the basis setup on an earlier one.
Throughout the module, I have gained the necessary knowledge into the revolutionary potential of machine learning models and how they will reshape the human future. I feel confident building similar networks that can be implemented in real-world situations by knowing the whole process and having worked on it in both individual and group settings. The knowledge of the weaknesses of the field and the potential threats either economical, physical or ethical will surely play a key role to motivate further research from my side to prevent them.
Miroslav, K. (2021) An Introduction to Machine Learning. 3rd Ed. Springer., 3(5), pp.2-4. DOI: https://link-springer-com.uniessexlib.idm.oclc.org/book/10.1007/978-3-030-81935-4
Patil, S., Nagaraja, G. (2020). Exploratory Data Analysis. International Research Journal of Engineering and Technology (IRJET) 7(05). DOI: https://www.academia.edu/download/64615158/IRJET-V7I51256.pdf [Accessed 20 January 2025].
Hutson, M. (2021) Robo-writers: The rise and risks of language-generating AI. Nature, 591(7848), pp.22-25. DOI: https://www.nature.com/articles/d41586-021-00530-0
Tohidul, M. & Nafix, K. (2018). Image Recognition with Deep Learning. IEEE 22(5). DOI: https://ieeexplore.ieee.org/abstract/document/8550021 [Accessed 6 January 2025].
Kaggle (2008). CIFAR-10 Object Recognition in Images. Available via: https://www.kaggle.com/competitions/cifar-10/overview [Accessed 6January 2025].
Keiron, O., Nash, R. (2015). An Introduction to Convolutional Neural Networks. Available via: https://arxiv.org/pdf/1511.08458 [Accessed 23 January 2025].
Mumuni, A., Fuseini, M. (2022). Data augmentation: A comprehensive survey of modern approaches. AI Journal 16(1). DOI: https://www.sciencedirect.com/science/article/pii/S2590005622000911 [Accessed 24 January 2025].
Nazri, M., Atomi, W. (2014). The Effect of Data Pre-processing on Optimised Training of Artificial Neural Networks. Procedia Technology 11(1): 32-39. DOI: https://www.sciencedirect.com/science/article/pii/S2212017313003137 [Accessed 24 January 2025].
Diez, A., Ser, J. (2019). Data fusion and machine learning for industrial prognosis: Trends and perspectives towards Industry 4.0. Information Fusion 10(50): 92-111. DOI: https://www-sciencedirect-com.uniessexlib.idm.oclc.org/science/article/pii/S1566253518304706?via%3Dihub [Accessed 25 January 2025].
The Group project required a submission of an analytical report discussing the process of performing exploratory data analysis and data pre-processing on the Airbnb dataset. The business question chosen was {}, and the following submission was made:
The Individual project was to develop an artificial neural network designed to perform image classification tasks. The following submissions were made: