Special Sessions

The quest for commercial fusion power is gaining momentum thanks to the use of artificial intelligence (AI) in nuclear fusion research. The process of nuclear fusion, which mimics the reaction happening in the sun, has the potential to produce energy that is safe, clean, and nearly infinite. The interest in fusion has risen recently, with companies racing to deliver commercial fusion and increased collaboration with government programs. Despite the promising advances, significant challenges remain. The path to commercial fusion will require massive investments and an expanded workforce of skilled researchers and engineers. International collaboration is also essential to share knowledge and experimental techniques, pushing fusion research closer to practical deployment. 

In the era of deep learning, where neural networks are widely deployed across diverse applications, the demand for scalable, efficient models has never been higher. This special session on Neural Network Compression Techniques aims to address the growing challenges of computational overhead, memory constraints, and energy consumption in deploying deep learning models on resource-limited devices and in distributed environments.

Biomedical data analysis involves the treatment of the physiological electrical activities measured using sensors placed on a living thing, also the medical imaging, allowing it to provide a useful process for abnormality detection and diagnosis purposes. Recently, Deep Learning (DL) has received a great attention to solve difficult and complex problems related to biosignal and medical image processing, where the traditional signal/image processing algorithms and conventional machine learning techniques have shown their limitations to solve such problems. 

The proposed special session, "Advances in Time-Series Data: Novel Theories, Methods, and Applications," aims to address the rapidly growing research area of time-series data analysis, which has profound implications across a wide array of fields, including finance, healthcare, environmental science, and engineering. Time-series data is critical for understanding and predicting patterns over time, yet traditional methods often fall short in handling the complex, dynamic, and large-scale nature of modern time-series datasets.

The proposed special session, "Advancing Physics-Informed Neural Networks (PINNs): Bridging Scientific Principles and Machine Learning for Complex Systems," aims to explore the latest developments, challenges, and applications of PINNs, a rapidly evolving paradigm at the intersection of computational science, physics, and artificial intelligence. PINNs leverage the synergy between data-driven approaches and physical laws, embedding domain knowledge into neural networks to solve complex problems in scientific computing.

This special section seeks to explore the transformative role of neural networks, in advancing the design, analysis, and resilience of engineering structures. Neural networks, with their ability to model complex, nonlinear relationships, have become indispensable tools in structural engineering. They are being increasingly integrated into engineering workflows, driving innovation in structural design, safety evaluation, and predictive maintenance. 

The healthcare domain is undergoing a transformative revolution driven by the adoption of artificial intelligence (AI) technologies, among which Foundation Models (FMs) are emerging as game changers. Foundation Models, encompassing large-scale pre-trained models like GPT-4, BERT, and others, offer a unified approach to dealing with a multitude of tasks by leveraging extensive pretraining on diverse datasets. 

The potential of AI extends far beyond commercial applications, with a profound capability to address pressing societal challenges. From mitigating climate change and improving healthcare access to reducing poverty and enhancing education, AI has the power to create tangible benefits for humanity. 

Breakthroughs in Artificial Intelligence (AI) are increasingly permeating everyday life, reshaping sectors such as the economy, medicine, military, education, and creative industries. As global powerhouses strive to foster innovation and attract investment, they are also enacting regulations to safeguard the interests, rights, and freedoms of their citizens. Consequently, the ethical and regulatory challenges across the AI value chain—spanning design, development, and deployment—are now at the forefront of legal and policy debates.

The special session aims to explore the challenging relationship between Artificial Intelligence (AI) and complex systems through Complexity Theory, delving into the understanding of modern AI systems such as deep learning architectures and large language models (e.g., GPT/ChatGPT, Gemini, Llama, Claude, etc.) as complex dynamic systems. 

The field of neuroscience is experiencing a data revolution, driven by advances in recording and imaging technologies that generate complex and high-dimensional neuronal datasets. 

In recent years, the European Commission has significantly increased its investment in Artificial Intelligence (AI) and Machine Learning (ML) research as part of its strategy to position Europe as a global leader in digital innovation.

The proposed special session will serve as a premier platform for researchers, practitioners, and industry experts to present innovations, share insights, and discuss future trends in the rapidly evolving intersection of artificial intelligence (AI) and neural engineering. 

Artificial intelligence transformer-based models such as the BERT model (Bidirectional Encoder Representations from Transformers), GPT (Generative Pre-trained Transformer), vision transformers (ViT) and many others have found significant applications in healthcare. 

The growing complexity of software systems and the constant need to adapt them to the changing needs of users and markets make constant software evolution essential. Therefore, the need to ensure quality and adaptability to software systems makes it urgent to integrate advanced solutions that can assist development teams in the evolutionary and maintenance process. 

The development of intelligent systems capable of reasoning and learning under uncertainty is fundamental in several contexts. The Bayesian perspective provides a powerful framework for modeling complex dependencies and handling incomplete information, establishing it as an essential tool in various fields. 

Prof. Zoubin Ghahramani said in his Nature paper, “intelligence relies on understanding and acting in an imperfectly sensed and uncertain world”. 

Such uncertainty handling ability is even more critical for some safety-critical tasks (e.g., autonomous driving and medical diagnosis). Unfortunately, existing neural networks are weak on that. The topic of this session - Bayesian neural networks - is to combine the beauties of two fields: neural networks which is powerful on complex function approximation and hidden representation learning, and Bayesian which has solid theoretical foundation on uncertainty modeling. 

The aim of this special session is to bring together research that advances the integration of biologically inspired neural networks and learning systems for robotics and mechatronics. Biologically inspired neural models, such as neural dynamics, have shown promising applications in solving complex problems in robotics, including motion planning, localization, navigation, and mapping. 

In an era dominated by data-driven insights, designing and building trustworthy Computational Intelligence (CI) systems presents fundamental challenges, especially when leveraging data from decentralized sources. Privacy regulations and ethical concerns have raised the demand for CI systems that are both collaborative and reliable, all while upholding stringent standards of data privacy and transparency. 

Complex-valued and, more generally, hypercomplex-valued neural networks (HVNNs) constitute a rapidly growing research area that has attracted continued interest for the last decade. Besides their natural ability to treat multidimensional data, hypercomplex-valued neural networks can benefit from hypercomplex numbers’ geometric and algebraic properties. 

In recent years, while the field of Software Engineering (SE) has seen the development of innovative methodologies and frameworks, their broader technological transfer has continued at a gradual pace. This slow uptake is primarily due to the unique challenges within SE as compared to more established engineering disciplines. 

With the increasing penetration of distributed energy resources, such as electric vehicles (EVs), photovoltaics (PVs) and prosumers, it becomes more challenging for the central system to balance the demand and supply in real-time. 

As neural network techniques progress, their uses increasingly transcend disciplinary borders, crafting a fertile ground for innovation. This special session, "Cross-Domain Innovations in Neural Network Methods and Applications," will highlight groundbreaking research utilizing neural networks across various sectors, including healthcare, finance, climate science, and quantum computing. This session aims to facilitate knowledge sharing and inspire new neural network designs to tackle complicated, real-world issues by bringing together experts from different fields. 

Internet of Things (IoT) is facilitated by heterogeneous technologies, which contribute to the providing of innovative and intelligent services in large number of application domains. 

The digital transformation is inextricably influencing our lives today. The constant flow of data and technological innovations have reshaped the way we live. The advent of technologies such as cloud computing, the Internet of Things, and Artificial Intelligence (AI) has given rise to an interconnected digital ecosystem, where billions of devices generate an enormous amount of data and information. 

Vision Transformers (ViTs) have revolutionised computer vision, offering state-of-the-art performance across a range of tasks. However, their heavy dependence on large datasets poses significant challenges, particularly in scenarios where data collection is constrained due to cost, privacy, or resource limitations. 

The goal of this event is to bring researchers in machine learning, AI, distributed systems, and electrical and electronic engineering together from across the globe. With the breakthroughs in Deep Learning (DL), recent years have witnessed the booming of Artificial Intelligence (AI) applications and services. 

Digital twin models are recently used in variety of research areas where they help to reduce operation costs and prevent possible malfunctions of real world devices. Deep learning models are constantly growing in new aspects, both theoretical and practical, important for various applications in modern technology and industry. 

In the last decade, the state-of-the-art has seen rapid advancements in designing and training powerful deep Neural Networks (NNs), especially Convolutional NNs, Graph NNs, Variational Autoencoders, Diffusion Models, Transformers, and Large Language Models (LLMs). Applying these AI models to biomedical data has opened unprecedented opportunities in the field of computational biology, revolutionizing both the traditional drug development paradigm and the way to analyze the huge amount of biomedical data coming from multi-omic approaches, which are needed to unravel the underlying mechanisms of diseases. In particular, drug discovery has become increasingly popular after AlphaFold. Drug discovery is a complex process aimed at identifying molecules that can interact with specific molecular targets to treat diseases. 

Generative Adversarial Networks (GANs) have become new hotspots of Artificial Intelligence. Various GAN-inspired and variant models (e.g. conditional GAN (cGAN), auxiliary classifier GAN (AC-GAN), bidirectional GAN (BiGAN), semi-supervised GAN and variational autoencoder GAN (VAE-GAN)) have been developed for image generation, segmentation, detection and classification, as well as video/audio synthesis/retrieval/restoration/domain adaptation. Such developments have contributed to a variety of applications, such as personalized story-telling, animated movie generation, building/interior design, e-learning, medical diagnosis, surveillance, agriculture, deepfake generation, urban land use detection, and underwater/forest observation using multispectral/hyperspectral images. 

Modern AI and advanced sensing technologies have been transforming our ability to observe the Earth and explore the Universe as well as monitor space objects. By analysing and interpreting multifaceted data (primarily imagery) captured by sensing devices on aircrafts, UAVs, satellites, spacecrafts, and astronomical telescopes that operate either on ground or in orbit, insights can be gained to enable intelligent decision-making with high autonomy. 

Graph structures allow to represent structured/complex data with entities and their relationships. These data structures naturally characterize a wide range of problems, including the areas of bio/chemistry (e.g. (macro-)molecules, proteins, biological networks), natural language processing (e.g. strings and parse trees), social network analysis (e.g. link prediction, object identification), information dissemination on social networks (temporal/dynamics graphs), and epidemiological studies (graph agent-based models). 

The advent of digital twins has revolutionized the simulation and optimization of real-world scenarios. Digital twins are comparable virtual replicas of real-world systems, assets, or processes that allow for real-time optimization, simulation, and monitoring. Through a complete or semi-complete digital replication of a physical object, they provide performance analysis, problem prediction, and scenario testing without affecting the real system. 

The scope of this workshop is to address the growing challenges of using distributed machine learning (ML) and artificial intelligence (AI) in Internet of Things (IoT) devices and edge computing systems. 

Transfer learning aims to leverage knowledge acquired from source models to tackle target tasks, even when the source and target data come from different distributions or modalities. 

A recurring concern in the use of neural network-based models is the opaqueness/lack of transparency and explainability as well as risk assessment. With the rise of generative AI and foundational models, specifically large language models and vision transformers, we need to ensure that our computational scientists are integrating AI ethics in every component and addressing security and privacy issues. 

Evolutionary computation is a computational paradigm inspired by the principles of biological evolution, and it has found applications in various fields, including wireless communications. Evolutionary algorithms simulate the process of natural selection, reproduction, and mutation to evolve solutions to optimization problems. 

The rise of AI in critical fields such as healthcare, autonomous systems, and finance demands greater transparency in decision-making processes. Neural networks, despite their high performance, remain opaque, creating a barrier to trust and accountability. 

Artificial intelligence is widely adopted in bioinformatics and bioengineering. As a matter of fact when the diagnosis or selection of therapy is no longer performed exclusively by the physician, but to a significant extent by artificial intelligence, decisions easily become nontransparent. The most common application of machine learning algorithms in the bioinformatics and bioengineering context is automatic clinical decision-making. 

In the era of data-driven governance, linked open government data (LOGD) plays a crucial role in enhancing transparency, accountability, and public trust. Artificial Intelligence (AI) has demonstrated its potential to extract insights from Open Data, enabling policy-makers and citizens to make informed decisions. However, the opaque nature of many AI models poses significant challenges to their adoption in government and civic contexts, where explainability is paramount. 

The adoption of Deep Neural Networks (DNNs) in critical domains such as healthcare, governance, and misinformation detection has created an unprecedented demand for high-performance and interpretable models. Despite their success, DNNs' opaque nature undermines user trust and raises ethical concerns, especially in sensitive applications. 

The rapid development of machine learning (ML) has led to a wide range of applications in various fields, including healthcare, finance, and natural language processing. However, the success of ML models often relies on extensive expertise in feature engineering, model selection, and hyperparameter tuning. To address this issue, Automated Machine Learning (AutoML) has emerged as a promising approach to democratize AI by automating these complex processes. 

Foundation Models (FMs) represent a significant advancement in deep learning architectures, characterized by their large-scale nature and task-agnostic design. Trained using self-supervised learning on extensive, diverse datasets, these models acquire a generalized understanding applicable across a spectrum of downstream tasks. 

The rapid advancement of generative AI technologies has brought both transformative potential and unprecedented challenges in the realms of privacy and security. 

In Artificial Intelligence, there is an increasing demand for adaptive models capable of dealing with a diverse spectrum of learning tasks, surpassing the limitations of systems designed to tackle a single task. 

Graphs are particularly well-suited for representing interacting systems, as they excel at capturing structural dependencies. This versatility has made graphs a powerful tool for modeling a wide range of phenomena, from the structures of molecular compounds to the dynamic interactions within social networks, as well as the complexities of transportation systems.

Graph data is prevalent across diverse fields such as social networks, bioinformatics, transportation systems, recommendation engines, and communication networks. These complex data structures inherently capture intricate relationships and interdependencies that traditional artificial intelligence methods struggle to fully leverage due to their structural complexity. 

Human-centered Artificial Intelligence (AI) represents a paradigm shift in the development and deployment of AI technologies, emphasizing the augmentation, complementarity, and enhancement of human capabilities, rather than replacing them. This approach intertwines the technical prowess of AI with ethical, social, and human values, ensuring that AI advancements are not only effective but also beneficial and accessible to all segments of society. 

The special session will focus on the current research trends in theory and practical aspects of systems that enhance and complement Computational Intelligence (CI) and Artificial Intelligence (AI) technologies to design and develop cognitive architectures enabling learning and abstraction abilities, and intelligent systems capable of human-like language understanding.

The Integrated Machine Learning and Wireless Communication (IMAC) special session at IJCNN 2025 aims to address the critical convergence of machine learning (ML) and wireless communication, which is a nexus pivotal to future AI applications and the next-generation networks. 

The research and development of intelligent vehicles and transportation systems are rapidly growing worldwide. Intelligent transportation systems are making transformative changes in all aspects of surface transportation based on vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I) connectivity, and automated driving (AV). 

Recent years have witnessed remarkable advancements in foundation models in various domains. Trained on large-scale diverse datasets in a self-supervised manner, these models serve as a basis for various downstream tasks, offering adaptability and robustness. 

Large Language Models (LLMs) demonstrate a significant breakthrough in artificial intelligence (AI), revolutionizing wide range of fields such as machine learning (ML), human computer interaction (HCI), natural language understanding (NLU), natural language generation (NLG), content generation and personalization, domain-specific code generation. 

In emerging applications such as robotics, autonomous vehicles, and smart cities, Large Language Models (LLMs) are increasingly being explored as key components for enabling real-time decision-making in navigation and control tasks. 

Over the last decades there has been an increasing interest in using machine learning and in the last few years, deep learning methods, combined with other vision techniques to create autonomous systems that solve vision problems in different fields. This special session is designed to serve researchers and developers to publish original, innovative and state-of-the art algorithms and architectures for real time applications in the areas of computer vision, image processing, biometrics, virtual and augmented reality, neural networks, intelligent interfaces and biomimetic object-vision recognition. 

While there has been a lot of research into the use of optimization for improving the performance of machine learning approaches, research into machine learning for optimization is in its infancy. This special session aims to promote research in this area for further advancement of the field. 

The recent trends in energy systems evolution towards to sustainable production and use of energy employ information, communication, and automation technology to deploy an integrated power system with sustainability challenges in the complete value chain, from power generation, to transmission, distribution, final users and stakeholders. 

Multimodal deep learning is a dynamic and innovative research area in a neural network. It focuses on developing sophisticated models with new types of architectures and mechanisms. These models enable the creation of intelligent systems capable of extracting knowledge from different types of data leading to groundbreaking advancements in neural networks.

Deep neural networks have demonstrated substantial promise in a wide range of real-world applications, primarily due to their intricate architectures developed by domain experts. However, the architectural design process is often labor-intensive, which has imposed significant constraints on the further advancement of deep neural networks. 

The landscape of information retrieval (IR) and recommender systems (RecSys) has been fundamentally reshaped by advances in neural methods.Leveraging deep learning, transformational architectures, and advanced neural embeddings, the IR and RecSys fields are witnessing rapid innovations that enable more personalised, context-aware, and accurate recommendations and retrieval results.

Data-driven models and Machine Learning (ML) methods can unleash the full potential of nondestructive evaluation (NDE) and structural health monitoring (SHM), making them key technologies for achieving various objectives of industry 4.0 and 5.0. Machine learning can improve the efficiency of the classification of NDE data, which is still frequently done by trained operators, by supporting their decisions, or replacing them in some applications facilitating automation of nondestructive evaluation analysis. 

This special session aims to bring together researchers and practitioners in the fields of neuromorphic computing, spiking neural networks, and autonomous systems (e.g., robotics). With the growing demand for intelligent, adaptive, and efficient autonomous systems across various application use-cases such as robotics, mobile agents (e.g., UAVs), and self-driving vehicles, neuromorphic computing offers a bio-inspired approach to enable these systems with embodied intelligence. 

Human Pose Estimation (HPE) has rapidly become a cornerstone of AI-driven computer vision, employing advanced neural networks to interpret human postures and movements from visual or sensor data. HPE is a key technology in healthcare, security, sports analytics, workplace ergonomics, and human-computer interaction applications. 

Privacy-Preserving Machine and Deep Learning (PP-MDL) is an emerging research area focused on enabling inference and training of Machine and Deep Learning (ML and DL) models in ways that protect the privacy of user data, often originating from as-a-service platforms. This interdisciplinary field spans Artificial Intelligence (AI), statistical methods (such as Differential Privacy (DP) and k-Anonymity), and cryptographic techniques (such as Homomorphic Encryption (HE) and Multi-Party Computation (MPC)). 

In recent years, quantum computing has witnessed significant advancements, evolving from its conceptual foundations in the 1980s to the development of hardware prototypes in the 2020s capable of operating with hundreds of qubits. Despite its early-stage maturity, rapid progress in quantum hardware and algorithm development has ignited discussions on the potential computational supremacy of Noisy Intermediate-Scale Quantum (NISQ) devices over classical systems. 

Randomization-based learning algorithms have received considerable attention from academics, researchers, and domain workers because randomization-based neural networks can be trained by non-iterative approaches possessing closed-form solutions. Those methods are generally computationally faster than iterative solutions and less sensitive to parameter settings. 

Over the past decade, the Internet of Things (IoT) has become one of the most influential technologies in the fields of wireless communications and mobile computing. Originated from RFID and wireless sensor networks (WSNs), the paradigm of IoT has been transforming every aspect of human life including healthcare, energy, transportation, and manufacturing. 

Reservoir Computing (RC) has emerged as a computationally efficient approach to training recurrent neural networks (RNNs), offering unique capabilities in handling temporal data without the high resource demands of traditional deep learning models. With the increasing need for low-power, fast, and adaptable neural networks across domains like real-time signal processing, edge computing, and neuromorphic systems, RC presents a compelling solution. 

With the fast development of machine learning algorithms, foundation models (FMs) have been proposed to address multiple tasks at the same time. A well-known example is ChatGPT which has significantly impacted various aspects of our daily lives. 

The growth of neural network applications requires to develop risk management models that can balance opportunities with risks. In the session, we contribute to the development of neural network risk models presenting a set of integrated statistical metrics that can measure the ``Sustainability'', ``Accuracy'', ``Fairness'' and ``Explainability'' of any Artificial Intelligence application, in line with  the requests of the European Artificial intelligence act. 

The motivation for this special session stems from the need to explore and advance the field of continual learning, a critical area in artificial intelligence that seeks to emulate the human ability to learn from experiences in an ongoing manner. The continual structuration of information and knowledge is essential for the development of intelligent systems capable of adapting to new environments, learning from their own experiences, and acquiring more complex knowledge over time. 

Artificial Intelligence (AI), specifically Neural Networks (NNs), is transforming the landscape of pervasive computing, connectivity, and intelligence across various applications. Integrating AI into Internet-of-Things (IoT) and communication networks is essential for the future of sectors such as intelligent transportation systems, robotics, and telemedicine. 

Quantum Machine Learning (QML) is an emerging interdisciplinary field at the intersection of quantum computing and machine learning. This is an area of growing interest because QML exploits the computational advantages of quantum mechanics to potentially develop more efficient algorithms capable of solving problems that are infeasible for classical methods. 

The application of dynamical systems theory to enhance learning efficiency has gained traction due to its ability to improve explainability, interpretability, and provide provable guarantees. By offering a structured framework to analyze information flow, stability, and convergence, dynamical systems theory strengthens deep learning models, enabling better performance and robustness. This perspective transforms optimization into a control problem, streamlining processes such as hyperparameter tuning and convergence analysis. 

The computing everywhere paradigm is paving the way for the pervasive diffusion of tiny devices endowed with intelligent abilities (such as Internet-of-things and edge computing devices). 

The motivation for this special session on Federated Learning (FL) at IJCNN 2025 stems from FL's transformative role in enabling collaborative machine learning across industries while safeguarding user privacy and data security. FL's applications in healthcare, finance, telecommunications, smart cities, retail, automotive, and education demonstrate its versatility and value in leveraging decentralized data to achieve personalized services, fraud detection, network optimization, autonomous driving, adaptive learning, and more. By reducing data transfer, democratizing AI, ensuring regulatory compliance, and minimizing bias, FL has positioned itself as a crucial tool in today's data-driven world. 

The emergence of the big data era in healthcare has led to the widespread collection of health information in various forms, including electronic patient records, administrative claims data, biometric data, sensor data, and medical images. 

Federated Learning (FL) has emerged as a groundbreaking approach to enabling AI across decentralized and sensitive data sources without requiring data centralization, addressing privacy concerns that align with regulations like GDPR. For a leading conference like IJCNN, which emphasizes advancements in neural networks and transformative AI, FL offers a rich avenue for exploration.