Research
Research interests, publications, and ongoing academic work
Research Interests
Continual Learning
Building models that learn new tasks sequentially without catastrophic forgetting — my core research area, including online continual learning strategies, benchmarks, and evaluation protocols.
Deep Learning
Designing and analyzing neural network architectures and training strategies, with a focus on transfer learning and feature fusion across spatial and frequency domains.
Computer Vision
Medical image analysis including brain tumor classification and segmentation from MRI, combining CNNs, transfer learning, and wavelet-enhanced features.
Neuroimaging
Applying continual learning to neuroimaging tasks such as tumor segmentation, modality registration, and disease classification, with standardized benchmarks.
Pattern Recognition
Statistical and learning-theoretic foundations of recognition systems, from classical methods to modern deep representations.
Multimodal Learning
Emotion recognition through speech processing and virtual reality, integrating affective computing with multimodal frameworks.
Publications
Online Continual Learning: A Systematic Literature Review of Approaches, Challenges, and Benchmarks
Seyed Amir Bidaki, et al.
Neural Networks (Elsevier)
Show Abstract
A systematic literature review on online continual learning, analyzing over 2,000 publications and extracting detailed data from 81 key articles after quality assessment. The study covers strategies, learning settings, evaluation metrics, benchmarks, datasets, and quality attributes, resulting in a 46-page manuscript. As first author and main initiator, I contributed to all phases — from idea conception and methodology design to analysis and writing.
Continual Learning in Neuroimaging: A Comprehensive Survey and Benchmark Study
Seyed Amir Bidaki, et al.
In preparation
Show Abstract
Extending continual learning research to neuroimaging applications — tumor segmentation, modality registration, and disease classification. The work systematically analyzes existing approaches by model components, evaluation metrics, datasets, and methodologies, and proposes standardized benchmarks and continual learning settings such as domain-incremental segmentation.
Intrusion Detection by Leveraging Online Continual Learning
Seyed Amir Bidaki, et al.
In preparation
Show Abstract
Applying online continual learning to network intrusion detection, enabling models to adapt to evolving cyber threats without catastrophic forgetting. Investigates anomaly detection strategies, model architectures, and evaluation protocols for robustness and real-time adaptability in dynamic network environments.