My journey

From Mathematics to Artificial Intelligence

At secondary school, I chose the mathematics stream, and I had already developed a strong interest in mathematics. During my university studies, this interest not only remained strong but gradually became one of the central foundations of my academic and research career.

My undergraduate degree was in Software Engineering and Applied Mathematics. The curriculum combined applied mathematics, electrical and electronic engineering, computer science and software engineering. Calculus, engineering statistics and probability, differential equations, engineering mathematics and numerical methods formed an important part of our studies, alongside subjects such as electrical circuits, logic circuits, computer architecture, microcontrollers and microprocessors.

We also studied core computer science subjects, including the fundamentals of algorithms and programming, object-oriented programming, data structures, algorithm design, information storage and retrieval, databases, software engineering, operating systems, machine language, and the theory of languages and automata. I completed my undergraduate degree as the top-ranked graduate of my cohort.

After my bachelor’s degree, I continued with a master’s in Artificial Intelligence. For me, this was an entry into a very different world. In software engineering, we typically analyse a problem and design an engineered solution to address it. In artificial intelligence, the questions take a different form: how can we build systems that learn from data, discover patterns, make decisions, or exhibit intelligent behaviour?

During my master’s degree, I explored a range of fields, including machine learning, artificial neural networks, signal and image processing, evolutionary optimisation, swarm intelligence and fuzzy systems. My main focus was on combining neural networks with fuzzy logic. I wanted to bring together the strengths of both approaches: the ability of neural networks to learn from data and adjust their parameters, and the ability of fuzzy logic to represent knowledge and relationships through interpretable rules.

At the same time, I investigated how gradient-free optimisation algorithms could be used to improve the performance of neuro-fuzzy systems. For me, this was not simply a matter of combining two techniques. The central question was how their different capabilities could be integrated within a unified framework.

Immediately after my master’s degree, I began a PhD in Artificial Intelligence at the Science and Research University, Tehran. My doctoral research focused on cultural algorithms and jobs interaction theory. The core idea was to use neuro-fuzzy systems to generate components of a cultural algorithm, allowing the two systems to interact iteratively within a joint optimisation process.

Within this framework, each system could contribute to improving the other, with the aim of enhancing the performance of the overall optimisation process. Alongside this work, I also explored the integration of reinforcement learning and fuzzy logic.

At both master’s and PhD level, one branch of my research focused specifically on multi-objective optimisation algorithms. I also applied this approach to define multiple objectives when training neuro-fuzzy networks.

From Algorithms to the Mathematics of Advanced Models

After my PhD, my focus gradually shifted towards the mathematics underlying machine learning. I wanted to understand algorithms not only in terms of their implementation and applications, but also through their mathematical foundations, probabilistic principles, optimisation structures and underlying assumptions.

As part of this journey, I examined how machine learning was taught at more than 40 universities around the world. I was interested in how foundational concepts were introduced, how different algorithms were sequenced, and how the connections between theoretical principles, mathematical formulation and practical implementation could be made clearer.

Meanwhile, I continued to explore different areas of machine learning, from classical methods and foundational algorithms to ensemble and boosting techniques, including algorithms such as XGBoost.

With the emergence of deep learning, the scope of my interests and research expanded further. Deep neural networks, attention mechanisms, transformers and large language models raised new questions about model architectures, representation learning, and the extraction of complex patterns from data.

Over time, generative models also became an important part of my research interests, including variational autoencoders (VAEs), generative adversarial networks (GANs) and diffusion models. Alongside these, I studied probabilistic models, Bayesian networks, graph-based methods for social network analysis, market basket analysis, recommender systems and digital twins.

At different stages of my career, I also focused particularly on transfer learning and domain adaptation. These areas address the challenge of transferring knowledge from one domain to another while maintaining model performance when the underlying data distribution changes.

During my time at Bangor University’s School of Ocean Sciences, applied problems in marine science and renewable energy also became an important part of my work, from passive acoustic monitoring of marine mammals to offshore wind farm site selection and an offshore wind digital twin. In these projects, my focus extended beyond selecting an appropriate model to include precise problem formulation, defining optimisation criteria, and finding solutions suited to the practical constraints of the data and the application.

Today, my interests continue to centre on the mathematics underlying advanced machine learning models, particularly large language models and diffusion models, alongside fuzzy logic, reinforcement learning and digital twins. What matters to me is not simply keeping up with new technologies. I want to understand why these models work, how they are structured, what their limitations are, and how they can be analysed, trained and applied more effectively.

One Passion, Many Different Forms

Looking back at my academic journey, it might appear to encompass a diverse collection of fields: software engineering, mathematics, artificial intelligence, optimisation, machine learning, generative models and ocean-related applications.

For me, however, these are not separate disciplines. Each offers a different perspective on a shared question: how can we formulate a problem precisely, understand its structure, develop an appropriate solution, and establish why that solution works?

Perhaps that is why, after all these years, I still find it difficult to choose just one of these fields and leave the others behind. Each has its own appeal; each has taught me something valuable, and each has helped shape the way I think and the professional path I have followed.

To me, these fields are like my children. Each has its own character, but I care deeply about them all.