A5032557719- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- AI in cancer detection
- Multimodal Machine Learning Applications
- Radiomics and Machine Learning in Medical Imaging
- Machine Learning and ELM
- Advanced Graph Neural Networks
- Bone fractures and treatments
- Artificial Intelligence in Healthcare and Education
- Advanced Image and Video Retrieval Techniques
- Medical Image Segmentation Techniques
- Lower Extremity Biomechanics and Pathologies
- Diabetic Foot Ulcer Assessment and Management
- Brain Tumor Detection and Classification
- Topic Modeling
Korea Advanced Institute of Science and Technology
2025
The Gleason score contributes significantly in predicting prostate cancer outcomes and selecting the appropriate treatment option, which is affected by well-known inter-observer variations. We present a novel deep learning-based automated grading system that does not require extensive region-level manual annotations experts and/or complex algorithms for automatic generation of annotations. A total 6664 936 needle biopsy single-core slides (689 99 cases) from two institutions were used...
To address the issue of catastrophic forgetting in neural networks, we propose a novel, simple, and effective solution called neuron-level plasticity control (NPC). While learning new task, proposed method preserves existing knowledge from previous tasks by controlling network at neuron level. NPC estimates importance value each consolidates important neurons applying lower rates, rather than restricting individual connection weights to stay close values optimized for tasks. The experimental...
In capsule networks, the routing algorithm connects capsules in consecutive layers, enabling upper-level to learn higher-level concepts by combining of lower-level capsules. Capsule networks are known have a few advantages over conventional neural including robustness 3D viewpoint changes and generalization capability. However, some studies reported negative experimental results. Nevertheless, reason for this phenomenon has not been analyzed yet. We empirically effect five different...
To address the issue of catastrophic forgetting in neural networks, we propose a novel, simple, and effective solution called neuron-level plasticity control (NPC). While learning new task, proposed method preserves knowledge for previous tasks by controlling network at neuron level. NPC estimates importance value each consolidates important \textit{neurons} applying lower rates, rather than restricting individual connection weights to stay close certain values. The experimental results on...
Deep neural networks, which employ batch normalization and ReLU-like activation functions, suffer from instability in the early stages of training due to high gradient induced by temporal explosion. In this study, we analyze occurrence mitigation explosion both theoretically empirically, discover that correlation between activations plays a key role preventing persisting throughout training. Finally, based on our observations, propose an improved adaptive learning rate algorithm effectively...