Vahid Seydi is a Research Fellow in the School of Ocean Science at Bangor University in Data Science (DS) and Machine Learning (ML). Prior to Bangor, Vahid was an Assistant Professor at the Department of AI at Azad University South Tehran Branch (Feb 2014 - Sep 2020) and was an award-winning lecturer (Oct 2010 – Feb 2014). He received a B.Sc.(2005) in software engineering, M.Sc. (2007) and PhD(2014) in AI, from the Department of Computer Science at Science and Research University, Tehran Iran. He has been awarded Global Talen endorsement from the UK Royal Society (2023); his current research fellowship(2020); a merit-based scholarship for attending the school of AI, Rome, Italy(2019); a full scholarship Award from Azad University(2010-2014); and KNTU ISLAB Research Fellowship (2007-2010). Throughout his studies, he consistently achieved grades above 18 out of 20 in nearly all modules, and I often secured the first-ranked student. furthermore, in Zillow’s home value prediction Kaggle competition, he has been in the top 2% among 3779 teams of data scientists (2017).

He possesses 15 years of extensive experience in diverse areas of Data Science (DS) and Machine Learning (ML). His expertise spans across a wide range of topics including regression, classification, retrieval, clustering, reinforcement learning, probabilistic graphical models, Gaussian process, recommender systems, social network analysis, association rule mining, and optimization methods. Throughout his career, he has worked with various models and data types, such as tabular data, text, image, video, and acoustic signals.

He believes that it is our responsibility to strive towards creating a better world for future generations. The issue of global warming stands as one of the foremost challenges facing humanity today where we can significantly mitigate its effects by implementing renewable energy sources. Machine Learning methods have the potential to address many of the challenges associated with data collected in the field of offshore renewable energy. In alignment with Bangor University's vision, which aims to foster a "sustainable world for future generations", he currently contributes his expertise in AI and ML to the sector of marine renewable energy.

Research Interests:

  • Deep Learning, Domain Adaptation, Generative Models
  • Explainable Machine Learning
  • Reinforcement Learning
  • Optimization

I am available for consultation on data-driven issues, proposals, and projects. If you require expertise in ML and DS or need assistance with data-driven initiatives, I would be delighted to provide my insights and support. Please feel free to reach out to me for any collaboration opportunities or inquiries.

  • PhD: Artificial Intelligence- (thesis: Job’s interaction theory to train hyper-parameters of the cultural optimization algorithm.)
    2014
  • MSc: Artificial Intelligence- (thesis: multi-objective optimization to train neural networks and neuro-fuzzy systems.)
    2007
  • BSc: Computer Software Eng. - (Concentrations: RUP methodology, database management, SQL, object-oriented programming, designing algorithm, data structure, Java, Visual C++.)
    2005

Supervision

I have supervised more than 50 undergraduate students in the field of software engineering and software development since 2010. I supervised 21 ML projects for post-graduate students at the master’s level and supervised 4 and co-supervised 2 PhD students since 2014. 

PhD thesis:

Supervisor :

  • Mahta Hassanpour (2023), Adversarial Domain Adaptation.
  • Hossein Hajibabaie (2023) Community Detection in Social Networks Using Probabilistic Graphical Models.
  • Mahid Saadati (2021) Digital Image Watermarking in the field of Shearlet Transform based on SVD and meta-heuristic Algorithms.
  • Yeganeh Madadi (2020) Multi-Source Domain Adaptation via Low-Rank and Sparse Representation.

Co-Supervisor:

  • Serveh Lotfi (2020) Rumor detection in Twitter social network based on analysis of conversation propagation graph
  • Pejman Gholamnejad (2020) Multi-Objective Optimization Evolutionary Algorithm using clustering estimation of distributions

Master thesis:

  • Ali Aminzadeh Gohari (2022), The Application of Generative Adversarial Network in Text Anomaly Detection.
  • Mina Ameripour (2022), Sentiment Analysis with Graph Convolutional Networks using Directed PMI.
  • Mona Solgi (2021), Graph CNN to analyse online shopping.
  • Ehsan Nasiri (2021), The Application of Generative Adversarial Network in Recommendation System Design.
  • Nima Mashhadizadeh (2021), Breast Cancer Histology Image Classification Using Deep Neural Networks.
  • Elnaz Baktash (2020), Designing a Deep Autoencoder for Machine Translation based on Statistical Machine Translation and Attention Mechanism.
  • Mehrdad Hosseini Naveh (2020), Improving Model-based Collaborative Filter Recommender system Usin User-Empeding, Item-Embedding and Deep Learning.
  • Saman Jamalabbasi (2020), Sentence simplification with deep reinforcement learning.
  • Sanaz Abaszadeh (2020), The Application of CNN in Sentiment Analysis.
  • Mehrdad Jannesar (2020), Text generation using LSTM Networks based on additive framework.
  • Hamideh Shooshtari (2020), Object detection using Deep YOLO algorithm based on additive framework.
  • Ali Mollaahmadi (2019), Adversarial image caption generator network
  • Arezoo Mirmahdi (2019), Community Detection in weighted Networks using BIGCLAM Algorithm.
  • Hassan Golshani (2019), Modeling Brain Activations using Hierarchical Latent Factors.
  • Ali Salmi (2019), Designing a Recommender System for Free Education Resources based on Users Activity in Social Networks.
  • Masoumed Nafari (2018), Using precision of users reviews to improve the performance of matrix factorisation method in recommender systems.
  • Elham Rajabian (2018), The Application of Deep Learning in Traffic flow prediction with big data challenge.
  • Hossein Raoof (2018), Heart Disorders Classification by heart sound signals using  Hidden Makoff model and Decision Tree.
  • Parvin Aghazadeh (2018), Trustworthy Recommender Systems based on the prevention of fake identities influence.
  • Aydin Abedinia (2018), Optimization of XGboost algorithm for semi-supervised learning.
  • Ehsan Hosseini (2018), The Application of Reinforcement Learning in Designing non-player Adaptive Agents in Computer Games.
  • Ehsan Hojatolahi (2018), Prunin Data to Increase the Performance of Support Vector Machine
   

Teaching

I was teaching a range of undergraduate and postgraduate modules in computer science and AI such as:

(The topics I covered are provided in the link to that module. However, I updated the modules’ content regularly and adopted best practices from leading universities worldwide to deliver the material effectively.)

PhD and Master's degree: 2014-2020

Bachelor's degree: 2010-2020

I have done most of my teaching in the Department of CS-AI at Azad University, South Tehran Branch, with which more than 41,000 students, is currently the second-largest university in Iran, and the largest in the country and middle-east in terms of technical and engineering fields, with around 10,085 students in the Faculty of Engineering.

My research interest topics which I have been actively involved in researching for at least the past three years include ML-Based Predictive Digital Twin using Probabilistic Graphical Models, Domain Adaptation, and few-shot learning, One-class classification model based on deep generative models, Reinforcement learning, Robust recommendation system and Social Network Analysis. My fascination lies in the mathematical concepts underlying these models, and I enjoy the challenge of formulating problems in their format while exploring innovative ways to make them scalable in big data environments.  Based on my experience in oceanography, the broad range of challenges that machine learning can address can be categorized into white-box data-driven modelling, where we have physical equations for modelling, but we need observation to calibrate and improve the model; Gray-box modelling, where we have incomplete physical equations, but we know some knowledge such as how phenomena affect each other; and black-box modelling, where modelling based on learning from data play a central role. In the field of black box modelling, one of the current challenges is the lack of transparency in powerful existing models. To address this issue, researchers are pursuing two distinct paths. The first involves the to evolve those machine learning models that are inherently explainable. The second path involves the utilization of methods that analyse and provide explanations for non-transparent models. I am particularly interested in exploring feature representation methods that shed light on the behaviour of these non-transparent models.

I am interested in collaborating as a supervisor/co-supervisor for PhD and MSc students, especially in the field of oceanography and marine renewable energy. My ongoing research focuses on these areas and includes interdisciplinary dissertations where data-driven modelling plays an important role. I believe that with my expertise and knowledge, I can make valuable contributions in this capacity. On the other hand, since learning from data has a strong mathematical background, the development of machine learning models itself is another field of interest to me. If you want to cooperate, especially in the following areas, please email me.

  • ML-Based Digital twin
  • Passive acoustic monitoring
  • Classification, Object detection, Object tracking (deep learning approaches)
  • ML topics like Generative models, domain adaptation, RL, social network analysis, recommender systems

2024

2023

  • PublishedA Survey on Adversarial Domain Adaptation
    HassanPour Zonoozi, M. & Seydi, V., 1 Jun 2023, In: Neural Processing Letters.
    Research output: Contribution to journal › Article › peer-review
  • PublishedCommunity detection in weighted networks using probabilistic generative model
    Hajibabaei, H., Seydi, V. & Koochari, A., 2 Feb 2023, In: journal of Intelligent Information Systems. 60, 1
    Research output: Contribution to journal › Article › peer-review
  • PublishedMachine learning in marine ecology an overview of techniques and applications
    Rubbens, P., Brodie, S., Cordier, T., Destro Barcellos, D., Devos, P., A Fernandes-Salvador, J., I Fincham, J., Gomes, A., Olav Handegard, N., Howell, K. L., Jamet, C., Heldal Kartveit, K., Moustahfid, H., Parcerisas, C., Politikos, D., Sauzède, R., Sokolova, M., Uusitalo, L., Van den Bulcke, L. & TM van Helmond, A. & 18 others, T Watson, J., Welch, H., Beltran-Perez, O., Chaffron, S., S Greenberg, D., Kühn, B., Kiko, R., Lo, M., M Lopes, R., Ove Möller, K., Michaels, W., Pala, A., Romagnan, J.-B., Schuchert, P., Seydi, V., Villasante, S., Malde, K. & Irisson, J.-O., 1 Sept 2023, In: ICES Journal of Marine Science. 80, 7
    Research output: Contribution to journal › Article › peer-review

2022

2021

2020

2017

2015

2014

2010

2008

2007

  • PublishedImproving the non-dominate sorting genetic algorithm for multi-objective optimization
    Seydi, V., Khanehsar, M. A. & Teshnehlab, M., 2007, 2007 International Conference on Computational Intelligence and Security Workshops (CISW 2007). p. 89-92 4 p.
    Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review
  • PublishedMulti objective optimization of ANFIS structure
    Seydi, V., Shoorehdeli, M. A., Sharifi, A. & Teshnehlab, M., 28 Nov 2007, Intelligent and Advanced Systems, 2007. ICIAS 2007. International Conference on. p. 249-253 5 p.
    Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review
  • PublishedTraining ANFIS structure with modified PSO algorithm
    Seydi, V., Shoorehdeli, M. A. & Teshnehlab, M., 27 Jul 2007, 2007 Mediterranean Conference on Control Automation. p. 1-6 6 p.
    Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review
  • Smart Efficient Energy Centre (SEEC)

May 2019 - May 2023 (Finished)

SEEC is an interdisciplinary “big data” research centre, working across low-carbon energy sectors. One of these sectors is Marine Renewable Energy. SEEC takes a pioneering approach in utilizing advanced engineering, computer science, and modelling to address significant challenges related to enhancing the sustainability of energy supply and utilization. The focus is on minimizing adverse environmental effects, particularly in reducing net carbon emissions. SEEC is part-funded by the European Regional Development Fund, administered through the Welsh Government.

As a researcher in Machine Learning and Data Science, my contribution to the SEEC project within the Marine Renewable Energy sector has been focused on investigating and leveraging advanced techniques to address data-driven key challenges. By integrating predictive modelling and optimization approaches, I've worked on cutting-edge ML algorithms to analyse large-scale data sets, uncovering valuable insights and patterns related to energy production, environmental impacts, and system performance while enhancing the sustainability of energy supply and utilization.

  • Ecological implications of accelerated seabed mobility around windfarms (EcoWind-ACCELERATE)

Aug 2022 - Aug 2026 (Active)

Accelerating the shift from fossil fuels, offshore wind farms are being rapidly developed. However, their impact on the seabed and marine ecosystem needs urgent attention. The presence of wind turbines and anchors alters sea currents, causing sediment movement and reduced water clarity. These changes affect the seabed shape, composition, biodiversity, and ecosystem services like fishing and carbon storage. Disruptions in prey species impact deep-diving predators, including protected seabirds. This project evaluates the combined effects of wind farm expansion and climate change on the seabed, quantifying implications for biodiversity, ecosystem services, habitats, and seabird-food interactions.

ML-based Digital Twin (DT) is one of my research interests. DTs have the potential to revolutionize decision-making across science, technology, and society. A DT is a collection of virtual information constructs that mimic the structure, context, and behaviour of a physical asset. These constructs are dynamically updated with data from its physical twin throughout its lifecycle, enabling informed decisions that yield value. Currently, I am working on a DT for the NERC EcoWind project, focusing on understanding, and predicting the impacts of windfarms on the physical and ecological functioning of coastal seas.

  • Morlais Demonstration Zone - Environmental Monitoring and Mitigation Programme

April 2022 – January 2024 (Active)

A tidal energy project located offshore Anglesey in Wales is set to advance the development of tidal power generation technologies. The project, known as Morlais, aims to establish grid connectivity and enhance the infrastructure for tidal power generation. The Welsh Government has emphasized the importance of environmental monitoring and mitigation package (EMMP)  for the project. This includes monitoring interactions with sensitive species and testing monitoring technologies, addressing critical data gaps and challenges in the tidal sector. These efforts contribute to the sustainable progress of the Morlais project and the wider advancement of tidal energy technology.

In this project, my role is to utilize machine learning to analyse and address data-driven challenges, focusing on the behaviour of sensitive species, such as marine mammals. Machine learning techniques can play a crucial role in analysing diverse and voluminous data sets, particularly in cases where labelled data is limited. By leveraging machine learning, we aim to gain a deeper understanding of the behaviour patterns of these species, facilitating the development of effective strategies within the EMMP framework.

  • MEECE R&D project APT wind farm constraints tool

01/04/2022 – 30/06/2023 (Active)

Producing risk maps for wind farm site selection in the Celtic Sea requires analysis and assessment of a range of GIS (Geographic Information System) data (e.g., seabed bathymetry and substrate, shipping route density, military danger and exercise areas, UK oil and gas infrastructure, UK wind, wave, and tidal designated areas), is a complex and challenging task.

In this project, I am working on a novel integrated evaluation framework based on the rule-based fuzzy inference system. This framework brings together (GIS) and spatial multi-criteria decision analysis, to provide a trustable decision tool for windfarm site placement. Embedding this framework in a fuzzy platform not only addresses uncertainty in the decision-making process but also enhances the interpretability of complex information, an essential requirement for this sector.

  • Development of a cetacean classifier to support the study of Passive Acoustic Monitoring data used in the marine renewables sector in Wales - Wales Data Nation Accelerator (Welsh Government) - Sprint Award 

Feb 2022 – May 2022 (Finished)

Tidal stream turbines have the potential to play a significant role in the generation of marine renewable energy. However, the risk of large marine mammals encountering these turbines has raised concerns among regulators and conservationists. Passive acoustic monitoring is an effective and established method to remotely monitor marine mammals in situ, over extended periods of time and at high frequency, allowing the capture of episodic events. Analysis of the large resulting datasets is a significant challenge.

In this project, I've analysed the effective application of two fundamental structures of deep neural networks, CNN and RNN, for auto-classifying delphinid clicks. I've illustrated how to tweak models' structure to study their abilities in analysing both the raw waveform and the spectrogram of delphinid clicks. Through the application of deep learning techniques, our research has demonstrated the feasibility of identifying dolphins based on their unique biological clicks, even in the presence of significant background noise. Building upon this discovery, our focus now shifts to developing robust tracking algorithms specifically tailored to different dolphin species. These algorithms will enable us to gain valuable insights into the behaviour patterns of these species, further enhancing our understanding of their ecological dynamics.

  • Predicting failures in medium voltage lines from a sequence of SCADA

Apr 2019 - Jun 2019 (Finished)

One of the goals of reliability is to identify and manage the risks around assets that could fail and cause unnecessary and expensive downtime. Organizations know it is important to identify areas of potential failures and rate them in terms of likelihood and consequence. ENEL distribution must manage a very complex reality, several control centres (STUX and STM Systems), more than 2200 primary substations (HV/MV) and more than 100000 remote-controlled secondary substations (MV/LV) In substation automation systems, SCADA performs the operations like bus voltage control, bus load balancing, circulating current control, overload control, transformer fault protection, bus fault protection, etc. The main idea investigated in this project was to apply predictive maintenance to the medium voltage lines using only SCADA events messages (each of which is coded as a unique string), in order to predict component failures in the distribution grid.

In this project, a diverse range of methods was employed to identify patterns and signals in the data, specifically focusing on classifying anomalies within SCADA events that result in faults in medium voltage lines. The initial phase involved a comprehensive analysis of the data to gain insights into how to frame the problem effectively. Subsequently, a combination of supervised and unsupervised techniques was utilized to uncover patterns and identify event sequences that were associated with anomalies. The problem was approached from two angles: sentiment analysis (supervised) and anomaly detection (unsupervised) using natural language processing (NLP) techniques. This was based on the notion that, when considering the sequence of SCADA events leading to faults or non-faults, it could be likened to a sentence composed of words. To extract meaningful information, various embedding methods were applied, including CNN for text classification, RNN-based structures such as RNN, LSTM, and GRU, as well as customized waveNet, VAE, and GAN models. To further enhance the predictive capabilities, attention mechanisms were employed to extract significant SCADA events that were crucial to the occurrence of failures. By aggregating the representations of these informative events, a more comprehensive understanding of the underlying factors contributing to faults was achieved.

  • Zillow Prize - Improving Zestimate Home Valuation Accuracy (Kaggle Competition)

October 2017-January 2018 (Finished)

In this project, I participated in the renowned Zillow Prize competition, which aimed to enhance the accuracy of Zillow's Zestimate home valuation. With a prize of 1.2 million dollars, this competition presented a significant opportunity for the data science community to make a substantial impact on the real estate industry. The winning algorithms had the potential to influence the home values of over 110 million properties across the United States.

During the competition, I worked diligently as part of a competitive environment, leveraging my expertise in DS and ML to develop innovative approaches for improving the Zestimate model. Collaborating with a diverse group of data scientists and researchers, I employed advanced techniques in data analysis, feature engineering, and model optimization to create predictive models with enhanced accuracy.

By exploring various algorithms, experimenting with feature selection, and refining the model architecture, I aimed to uncover insights and patterns that would contribute to superior home valuation predictions. Additionally, I employed rigorous evaluation methodologies to assess model performance and iteratively refine my approach.

My efforts in the Zillow Prize competition resulted in achieving a commendable rank of 71 out of 3,775 participating teams. This accomplishment demonstrates my ability to tackle real-world challenges, apply cutting-edge techniques, and deliver competitive results within a highly competitive setting.

  • Hybrid optimization of fuzzy neuro system - research project - KNTU ISLAB

Sep 2008 - Aug 2010 (Finished)

A fuzzy neuro system, also known as a fuzzy neural network or neuro-fuzzy system, is a hybrid computational model that combines the principles of fuzzy logic and artificial neural networks. It integrates the ability of fuzzy logic to handle uncertainty and imprecise information with the learning and adaptive capabilities of neural networks. Incorporating fuzzy sets, fuzzy rules, and fuzzy inference mechanisms into the neural network architecture, enables the model to handle uncertain or imprecise data, perform fuzzy reasoning, and make accurate predictions or decisions based on the learned patterns. ANFIS and LLNFS are two popular models in this category.

In this research project, my focus has been on developing a framework that facilitates the application of hybrid optimization methods for training fuzzy neuro systems. The framework I have been working on aims to harness the advantages of both gradient-based and evolutionary-based techniques to optimize the training process of fuzzy neuro systems. Hybrid optimization methods leverage the strengths of both gradient-based and evolutionary-based techniques, providing enhanced convergence, robustness, scalability, and flexibility in the training process. They are particularly valuable in complex optimization scenarios with multimodal non-convex landscapes, noisy data, and large-scale problems.

  •  Multi-objective optimization - research project - KNTU ISLAB 

Sep 2007 - Aug 2008 (Finished)

Multi-objective optimization problems involve finding solutions that optimize multiple conflicting objectives simultaneously. These problems are often challenging because improving one objective may lead to a degradation in others, creating a trade-off situation. This concept is known as the "Pareto optimality" principle. The goal is to identify a set of Pareto frontier solutions that represent the best trade-offs among the objectives. Various algorithms explore the solution space to find a diverse set of options, allowing decision-makers to choose the most suitable solution based on their preferences or requirements.

In this project, I conducted a thorough examination of evolutionary algorithms and their capabilities in addressing multi-objective optimization problems. Evolutionary algorithms, being population-based, possess the unique ability to identify the Pareto frontier.  These algorithms leverage the principles of natural evolution, including selection, reproduction, and mutation, to iteratively improve the population of candidate solutions. By employing various mechanisms such as fitness assignment, selection operators, and genetic operators, evolutionary algorithms are able to offer a balance between exploration and exploitation, to search and converge towards the Pareto-optimal solutions.

TECHNICAL SKILLS

  • Programming Languages: Python, Java, C, C++, MATLAB, R, ProLog.
  • Python Stack: PyTorch, Keras, TensorFlow, Gensim, NLTK, NumPy, Pandas, SiKit-Learn, SciPy, Scrapy, Matplotlib, Seaborn, ...
  • Databases Management: SQL
  • Writing: LATEX, Microsoft Word, Markdown, HTML
  • Others: AWS cloud platform, Git, GitHub, Software Development, RUP, Agile.