Research
Hybrid
Neuro-Symbolic AI
Realizing ultimate AI by combining several disciplines based on modularity
Devising accountable fair learning algorithms with adversarial regularization
Explainable Fair AI
Industrial
Applications of AI
Solving real-world problems such as cyber security, fault diagnosis, life logging, VQA, etc.
.
.
.
Ongoing Project
Enhanced Neuro-Symbolic VQA
with Logical Reasoning
Advanced deep learning approach that automatically generates responses to queries after image input
Integration of symbolic artificial intelligence similar to human communication processes with Neural artificial intelligence
Development of an explainable deep learning model
Integration of external memory with inferable deep learning models
Combination of a neural network model responsible for learning and inference with memory that stores memory elements
Research on differentiable memory for implementing mutual organic learning and memory
Graph data format and time-series data input, inference learning after storing in memory
Visualization of the model and memory's access, usage, and input/output process
Research on Feature Imbalance in Natural Language Processing based on GAN
Research on bias, one of the factors that degrade the performance of deep learning
Elimination of bias based on the features of adversarial learning in Natural Language Processing using GAN
Research on bias removal methods through generalization in the latent space
Research on feature disruption methods based on data pairs corresponding to features after latent vector extraction
Comparison with other algorithms and validation of the strengths and validity of the said algorithm
Deep Learning-based Cyber Security
Development of a phishing URL classifier based on deep learning (Microsoft Texception Net)
Phishing characteristics based on URL: Immediate disposal of the program after utilization (user information theft, reporting)
Need for the development of a defense system that utilizes domain knowledge
Collection of prior knowledge in cyber security, conversion to a learnable and inferable format, and optimization
Goal: Visual representation and symbolization of the collected phishing links
