Nomica Choudhry

Deepfake detectors are usually trained on a fixed collection of manipulation methods, although synthetic video generation continues to evolve after deployment. This paper presents CDD-RFLNet, a continual open-world framework that combines RGB, constrained-residual, temporal-change, and forensic-disagreement cues. A calibrated gate uses predictive energy, residual-prototype distance, and temporal disagreement to identify unfamiliar clips. When labeled examples from a new domain become available, a frozen stability teacher preserves earlier knowledge while a plasticity teacher learns the emerging manipulation. Reliability-weighted distillation transfers their shared evidence, and localized disagreement controls lightweight stage adapters. Prototype replay limits forgetting without retaining the full historical training set. Trained initially on FaceForensics++, the model is evaluated through a sequential protocol that introduces unseen manipulation families and datasets. CDD-RFLNet achieves 91.4% unknown-detection AUROC and, after four adaptation stages, 93.7% average AUC with a standard deviation of 0.4% and only 1.0 points of forgetting. These results show that a compact detector can recognize distribution change, learn of emerging attacks, and preserve earlier forensic knowledge.

Accurate brain tumor classification is critical for patient prognosis and treatment planning, yet manual interpretation of medical images like MRI is subject to variability. Deep learning has emerged as a powerful tool for this task. This review charts the evolution of deep learning architectures for brain tumor classification. We conducted a comprehensive literature review, focusing on the architectural progression from foundational Convolutional Neural Networks (CNNs) to modern attention-based Transformer models. Key datasets, evaluation metrics, and clinical challenges are synthesized. The review details the trajectory from early CNNs (e.g., AlexNet, VGG), which excelled at local feature extraction, to advanced variants like ResNet, U-Net, and DenseNet that improved performance and enabled segmentation-classification workflows. The paradigm then shifted to Vision Transformers (ViT, Swin Transformer) and hybrid models, which explicitly model long-range dependencies and global context, often achieving state-of-the-art results. Challenges such as domain shift, data scarcity, and the need for explainability (XAI) are persistent themes. While both CNNs and Transformers have demonstrated high accuracy, the current state-of-the-art often involves hybrid architectures that leverage the strengths of both. Future progress lies in developing generalizable, efficient, and trustworthy models through techniques like self-supervised and federated learning, multimodal data fusion, and the development of large-scale medical foundation models, ultimately aiming to empower clinicians and improve patient outcomes.

Video Surveillance Systems (VSSs) are used in a wide range of applications including public safety and perimeter security. They are deployed in places such as markets, hospitals, schools, banks, shopping malls, offices, and smart cities. VSSs generate a massive amount of surveillance data, and significant research has been published on the use of machine learning algorithms to handle surveillance data. In this paper, we present an extensive overview and a thorough analysis of cutting-edge learning methods used in VSSs. Existing surveys on learning approaches in video surveillance have some drawbacks, such as a lack of in-depth analysis of the learning algorithms, omission of certain methodologies, insufficient critical evaluation, and absence of recent learning algorithms. To fill these gaps, this survey provides a thorough examination of the most recent learning algorithms for anomaly detection. A critical assessment of the algorithms including their strengths, weaknesses, and applicability as well as tailored classifications of anomaly types for different domains are provided. Our study also offers insights into the future development of learning techniques in VSS, positioning itself as a valuable resource for both researchers and practitioners in the field. Finally, we share our thoughts on what we learned and how it can help with new developments in the future.