Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
project
3D Medical Image Analysis for Objective Assessment of Orbital Disease
Period: 2019 – 2021
Orbit segmentation and 3D image classification for objective diagnosis and severity assessment of Graves’ orbitopathy.
Techniques: Orbit Segmentation, 3D Image Classification
Related Paper: Neural Network-based Method for Diagnosing Graves’ Orbitopathy Using Orbital Computed Tomography (Scientific Reports, 2022)
Code:
Deep Learning Self-Assembly Framework for On-Device AI
Period: 2020 – 2022
A self-assembly framework for on-device deep learning, focused on hardware-aware NAS and real-time human activity recognition with multi-scale temporal sampling.
Techniques:
- Hardware-Aware Neural Architecture Search (related to LUT-NAS, IEEE Access 2023, and group-structure NAS, Electronics Letters 2022)
- Real-time Human Activity Recognition (related to LUT-NAS and MTS-Module, IEEE Access 2022)
- Multi-scale Temporal Sampling (related to MTS-Module)
Code:
Clinical Deep Learning–based Image Analysis for Gallbladder Polyp and Wall Thickening
Period: 2020 – 2021
Medical image analysis pipeline for polyp segmentation and wall thickening assessment in gallbladder ultrasound and CT images.
Techniques: Polyp Segmentation, Medical Image Analysis
TANGO — Target Recognition No-Code Neural Network Creation and Operational Framework
Period: 2022 – 2023
A no-code framework for end-to-end neural network creation and operation, covering object detection, hardware-aware NAS, and post-training quantization.
Techniques: YOLO, Hardware-Aware NAS, Post-Training Quantization, Docker
Code & Resources:
Emotion-Specialized Text-to-Video Retrieval
Period: Jan. 2024 – Apr. 2024
A vision-language retrieval system that aligns video and text representations along an emotion-aware axis.
Optimized Molecular Generation with Molecular Representations
Period: Mar. 2024 – Jun. 2024
Latent-space optimization for molecular generation using a VAE with Bayesian optimization.
Techniques: Variational Autoencoder (VAE), Bayesian Optimization, Latent Space Analysis
Materials:
Sustainable Learning Machines for On-Device Holistic Intelligence
Period: 2023 – Present
Building deployable on-device learning systems through on-device training, online continual learning, hardware-aware NAS, multimodal learning, and OOD detection.
Techniques:
- On-Device Training (related to my CVPR 2026 publication on CPS-Prompt)
- Online Continual Learning (related to MixER, IEEE Access 2024)
- Hardware-Aware Neural Architecture Search (related to GTSNet, EAAI 2023)
- Multimodal Learning and Out-of-Distribution Detection (related to MAND, EIML @ ICML 2026)
Code:
publications
Multi-Temporal Sampling Module for Real-Time Human Activity Recognition
Published in IEEE Access, 2022
A multi-temporal sampling module that captures activities at varying time scales to enable accurate, real-time human activity recognition.
Recommended citation: J. Park, W. Lim, D. Kim, J. Lee. (2022). "Multi-Temporal Sampling Module for Real-Time Human Activity Recognition." IEEE Access.
Download Paper
Achieving an Optimal Group Structure in a Neural Architecture Search
Published in Electronics Letters, 2022
A method to search for an optimal group convolution structure within neural architecture search, improving accuracy-efficiency trade-offs.
Recommended citation: W. Seo, J. Park, W. Lim, D. Kim, J. Lee. (2022). "Achieving an Optimal Group Structure in a Neural Architecture Search." Electronics Letters.
Download Paper
Neural Network-based Method for Diagnosing Graves’ Orbitopathy Using Orbital Computed Tomography
Published in Scientific Reports, 2022
A neural network pipeline for diagnosis and severity assessment of Graves’ orbitopathy from orbital CT scans.
Recommended citation: J. Lee, W. Seo, J. Park, W. Lim, J. Oh, N. Moon, J. Lee. (2022). "Neural Network-based Method for Diagnosing Graves' Orbitopathy Using Orbital Computed Tomography." Scientific Reports.
Download Paper
Efficient Human Activity Recognition Using Lookup Table-based Neural Architecture Search for Mobile Devices
Published in IEEE Access, 2023
A lookup table-based neural architecture search method that rapidly identifies efficient HAR models tailored to mobile device latency profiles.
Recommended citation: W. Lim, W. Seo, D. Kim, J. Lee. (2023). "Efficient Human Activity Recognition Using Lookup Table-based Neural Architecture Search for Mobile Devices." IEEE Access.
Download Paper
GTSNet: Flexible Architecture Under Budget Constraint for Real-Time Human Activity Recognition from Wearable Sensor
Published in Engineering Applications of Artificial Intelligence, 2023
A budget-aware flexible architecture for real-time human activity recognition from wearable sensors, balancing accuracy and on-device latency constraints.
Recommended citation: J. Park, W. Lim, D. Kim, J. Lee. (2023). "GTSNet: Flexible Architecture Under Budget Constraint for Real-Time Human Activity Recognition from Wearable Sensor." Engineering Applications of Artificial Intelligence.
Download Paper
MixER: Mixup-based Experience Replay for Online Class-Incremental Learning
Published in IEEE Access, 2024
A mixup-based experience replay strategy that mitigates catastrophic forgetting in online class-incremental learning by interpolating samples across past and current tasks.
Recommended citation: W. Lim, Y. Zhou, D. Kim, J. Lee. (2024). "MixER: Mixup-based Experience Replay for Online Class-Incremental Learning." IEEE Access.
Download Paper
Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge
Published in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
A patch-aware sparse prompting method with decoupled training that enables continual learning on resource-constrained edge devices.
Recommended citation: W. Lim, J. Lee, D. Kim. (2026). "Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge." IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
Download Paper
MAND: Modality-Aware Novelty Detection for Open-World Egocentric Activity Recognition
Published in ICML 2026 Workshop on Epistemic Intelligence in Machine Learning, 2026
A modality-aware novelty detection framework for open-world egocentric activity recognition that leverages multimodal cues to distinguish novel activities from known ones.
Recommended citation: H. Im*, W. Lim*, D. Kim. (2026). "MAND: Modality-Aware Novelty Detection for Open-World Egocentric Activity Recognition." ICML 2026 Workshop on Epistemic Intelligence in Machine Learning. (* Equal contribution)
talks
teaching
Teaching experience 1
Undergraduate course, University 1, Department, 2014
This is a description of a teaching experience. You can use markdown like any other post.
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.