[CSE7001/8001] Creative Software (2026 Spring)

Date:  03/06 (Friday)

Speaker: Dr. Sung-Ho Bae, Assistant Professor, Kyung Hee University

Title: Psychophysical Vision Models and Their Applications to Image Processing

Abstract: The human visual system (HVS) is considered a very complex nonlinear function, and active research has been carried out in psychophysics to reveal the characteristics of the HVS. This presentation first introduces our recent studies on mathematical modeling of the HVS characteristics. Next, we introduce various applications of the HVS models in computer vision/image processing problems. Specifically, we introduce our new foundation that HVS has different distortion sensitivity depending on the distortion types and texture characteristics. Based on our foundation, we developed a new image quality assessment (IQA) metric that shows significantly high correlations with perceived visual quality. Furthermore, we extended our IQA method to have desirable mathematical properties to be applied in convex optimization problems, such as valid metric properties, differentiability, and convexity which were possessed by MSE (Mean Squared Error). Finally, we applied our HVS models and IQA methods in practical image quality optimization problems, such as video coding, and super-resolution, resulting in better performance in the perspective of visual quality perception compared to existing MSE-based methods.

Date: TBD

Speaker: Dr. Sangmin Lee, Postdoctoral Researcher (appointed in May), UIUC

Title: Associative Learning for Multimodal Representation under Ambiguous Pair Problems

Abstract: In order to understand the world at a human level, machines need to learn and be aware of the relationships among multimodal data beyond a single modal one. However, the relationships of given multimodal data in real-world environments are not always enough or certain for machines to learn. For example, when it is difficult to obtain certain modal data, the number of multimodal data pairs for machines to learn can be limited. In addition, even if there are enough multimodal data pairs, their relationships can be mismatched sometimes, which may confuse machines. Such situations with limited and mismatched pairs can be considered to have ambiguous pair problems that hinder machines from learning multimodal relationships. Therefore, it is necessary to address the ambiguous pair problems in order to learn multimodal feature representation robustly even in real-world environments. This talk will present multimodal association approaches that can deal with ambiguous pair problems by compensating for the lack of paired information. The approaches cover multimodal representation learning and multimodal retrieval tasks which suffer from limited and mismatched pair problems.

Date:  TBD

Speaker: Dr. Chaoning Zhang, Assistant Professor, Kyung Hee university 

Title: Self-Supervised Learning and Robustness in Deep Neural Networks

Abstract: In the past decade, deep learning has dramatically pushed the performance frontiers in vision tasks like image recognition. Take image classification as an example, convolutional neural networks and recent ViT models have shown steady progress in the state-of-the-art performance on the benchmark ImageNet dataset. Unfortunately, such a remarkable success often relies on training the model on numerous samples with labels, which is cost-intensive and time-consuming. To this end, self-supervised learning (SSL) has recently become a very active research direction to realize unsupervised (label-free) representation learning, for which contrastive learning is a major framework with significant attention. On the other hand, there is a growing need to improve the model’s robustness against adversarial attacks. In this seminar talk, Dr. Chaoning Zhang will cover recent progress on SSL for data-efficient and robust deep learning in computer vision.

Date:  TBD

Speaker: Dr. Hak Gu Kim, Assistant Professor, Chuang-Ang University

Title: Towards Human-like Machine Perception   

Abstract: Recently, artificial intelligence (AI) and machine learning (ML) have shown outstanding performances in various multimedia applications. However, there is a question about whether AI/ML work like human. In this talk, I will present my recent research on machine perception for advanced image and video processing inspired by human visual perception. First, I will introduce AI models that quantitatively predict the perceived visual discomfort in viewing immersive contents. By considering the mechanism of human perception in building the machine perception model, the AI model can perceive the discomfort and sickness that people feel. Next, I will present recent AI models considering perception information for image and video processing. Finally, I will introduce how to mimic what human experts do in order to make the machine perception model perceive and act like humans.

Date: TBD

Speaker: Dr. Gwangmu Lee,    Postdoc Researcher, EPFL  

Title: The chronicle of software vulnerability detection

Abstract: With the threats to cybersecurity ever-increasing, prematurely detecting vulnerabilities in software systems is deemed as one of the promising ways to thwart exploitation attempts from attackers at the fundamental level. In this talk, we first briefly recap software vulnerabilities and move on to the historical evolution of vulnerability detection techniques (i.e., how it has evolved up until now). Finally, we go over the current most popular vulnerability detection technique--fuzzing--and discuss its possible future development.