Publications

2026

1. CVPR26

   

   AdaRadar: Rate Adaptive Spectral Compression for Radar-based Perception

   Jinho Park, Se Young Chun, and Mingoo Seok

   In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
2. CICC26

   

   PersASR: A 2.3-μJ/frame 96.8%-accurate TinyML Automatic Speech Recognition Processor with Data Augmentation-Driven Personalization in 16-nm FinFET

   Seunghyun Moon^*, Jinho Park^*, Chuan-Tung Lin, and 5 more authors

   In IEEE Custom Integrated Circuits Conference (CICC), 2026

2024

1. ECCV24

   

   UL-VIO: Ultra-Lightweight Visual-Inertial Odometry with Noise Robust Test-Time Adaptation

   Jinho Park, Se Young Chun, and Mingoo Seok

   In European Conference on Computer Vision (ECCV), 2024

   Abs HTML PDF

   Data-driven visual-inertial odometry (VIO) has received highlights for its performance since VIOs are a crucial compartment in autonomous robots. However, their deployment on resource-constrained devices is non-trivial since large network parameters should be accommodated in the device memory. Furthermore, these networks may risk failure post-deployment due to environmental distribution shifts at test time. In light of this, we propose UL-VIO – an ultra-lightweight (<1M) VIO network capable of test-time adaptation (TTA) based on visual-inertial consistency. Specifically, we perform model compression to the network while preserving the low-level encoder part, including all BatchNorm parameters for resource-efficient test-time adaptation. It achieves 36X smaller network size than state-of-the-art with a minute increase in error – 1% on the KITTI dataset. For test-time adaptation, we propose to use the inertia-referred network outputs as pseudo labels and update the BatchNorm parameter for lightweight yet effective adaptation. To the best of our knowledge, this is the first work to perform noise-robust TTA on VIO. Experimental results on the KITTI, EuRoC, and Marulan datasets demonstrate the effectiveness of our resource-efficient adaptation method under diverse TTA scenarios with dynamic domain shifts.

2022

1. AICAS22

   

   Bin-Specific Quantization in Spectral-Domain Convolutional Neural Network Accelerators

   Jinho Park, Jaewon Lee, Gain Kim, and 1 more author

   In IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2022

   Abs PDF

   Spectral-domain convolution engines can effectively reduce the computational complexity of convolution operations. In these engines, however, element-wise multiplications of the spectral representations dominate the multiply and accumulate (MAC) operations. In light of this, we propose bin-specific quantization (BSQ), which is to judiciously allocate varying bit width to each spectral bin in overlap-save. This allows efficient computation of the Hadamard product since the magnitude of the high-frequency components in image features is significantly smaller than that of the low-frequency counterparts. Using the statistics from spectral representations of feature maps, we also delineate methods for properly allocating bit precision to those spectral bins. When BSQ is applied, the average bit precisions of the arithmetic operators in spectral-domain convolvers, without the requirement of network re-training, were lowered by 24% (AlexNet), 20% (VGG–16), and 22% (ResNet–18) while having no significant reduction (< 1%) on classification accuracy on the ImageNet dataset.

2021

1. ISCAS21

   Link bit-error-rate requirement analysis for deep neural network accelerators

   Jaewon Lee, Gain Kim, Jinho Park, and 1 more author

   In 2021 IEEE International Symposium on Circuits and Systems (ISCAS), 2021

   Abs PDF

   In convolutional neural network (CNN) accelerators, the dominant power consumption is caused by the access of external data memory. In addition, power and area occupied by I/O interfaces maintaining low bit-error-rate, e.g., 1e-15, grow as the data rate increases. Considering the inherent error resilience of the inference process in machine learning applications, the requirement of error-free communication in the data-path is controversial. In this paper, a custom CNN accelerator integrating a channel emulator is designed by using an FPGA to analyze the effect of the BER of an I/O transceiver on the image classification accuracy. In order to implement a channel emulator, a digital-domain look-up-table (LUT)-based 12-tap FIR filter is employed to create inter-symbol interference (ISI), and a PRBS31 generator is used as a noise source. The implementation was evaluated by running the ImageNet dataset on the FPGA-based custom accelerator (Virtex Ultrascale+) implementing VGG-16. The results show that the BER up to 1e-4 in the memory access has a negligible impact on the inference accuracy.

2. IST21

   Outdoor environmental sensitivity test for the detection of biological aerosols

   Hyunsoo Seo, Kibong Choi, and Jinho Park

   Instrumentation Science & Technology, 2021

   Abs

   A real-time light-emitting-diode-induced fluorescence biological aerosol detection method is developed that enables real-time monitoring of airborne particles derived from organisms so the presence of these contaminants can be identified. To characterize the performance, a variety of background measurements describing different contamination levels and particulate concentrations controlled to closed environments were obtained and used as the input to a detection algorithm. Optimization was performed for various backgrounds. Additionally, detection performance per sensitivity level was determined using a confusion matrix so that the user can select sensitivities depending on the background. For 50 to 100 agent-containing-particles-per-liter-of-air target detection performance, an alarm rate of 100% was achieved, and the false-alarm rate was optimized to 0%. Applying these settings, the final sensitivities of the “high” and “low” modes were roughly twice and half of those of the “normal” mode, respectively. The final alarm thresholds of high and low sensitivities were set to 0.1 times and 10 times the reference values, respectively, enabling reliable operation in a variety of situations.

2020

1. IEEE JSEN20

   

   Signal encoder of real-time bio-aerosol counter using 280 nm UV-LED induced fluorescence

   Jinho Park, Young-Su Jeong, Hyunwoo Nam, and 1 more author

   IEEE Sensors Journal, 2020

   Abs PDF

   Real-time monitoring of bio-aerosols in the atmosphere has gained importance in both military and civilian security. As the first stage of defense architecture, a fluorescent particle counter is a compact, low-power, fast response sensor capable of giving a tentative warning. In this research, we developed a novel signal encoder for low power real-time fluorescent particle counter. Using a single 280 nm UV-LED light source and an acceleration nozzle, the particle counts are binned to three sizes: 0.84-2, 2-5, and 5-11 μm. Integrated fluorescence is measured in 308-420 nm. Optimization of pulse width thresholds on the fluorescence channel based on fluorescent particle detection accuracy is demonstrated. Proposed size distinguishing scheme achieves accuracy of 0.853 compared to 0.823 for accuracy of single bin counterpart. The adaptive window scheme improves the dynamic range of particle counts. Further, the count efficiency improved by 0.394 for input count rate of 104 sec-1 compared to fixed size window scheme. Counting performance comparison with an APS validates good linearity of R2 = 0.984. As the device showcases high dynamic range, linearity, and detection accuracy, we expect that it can serve as a useful tool in ambient environment monitoring, especially concerning airborne fluorescent particles.

2. IST20

   Development of a real-time handheld bioaerosol monitoring system using ultraviolet-light emitting diode induced fluorescence

   Young-Su Jeong, Jong Min Lee, Jinho Park, and 2 more authors

   Instrumentation Science & Technology, 2020

   Abs

   A handheld biological particle monitoring system that analyzes airborne biological particles in real time was designed, fabricated and its performance was investigated. An ultraviolet-light emitting diode (UV-LED) light source with a 280-nm wavelength was used. The design of an optical module that is composed of an ultraviolet source, a nozzle and flow controller, an ultraviolet cell, and a detection region was described. Through performance measurements, it was confirmed that scattering signals that corresponded to each particle size were measured, and that the proposed system effectively distinguished biological particles from non-biological components. It is expected that the proposed system can be applied to various biological particle and environment monitoring systems.

2021

1. KR

   A method for detecting biological particles

   Jinho Park, Kibong Choi, Hyunsoo Nam, and 1 more author

   Mar 2021

   Korean Patent

   Technology Transfer

   Technology transferred to Samyang Chemical Co., Ltd. for defense production and deployment for USD 10k.

2. KR

   A small fluorescence sensor device for detecting fine particles

   Kibong Choi, Yeoncheol Ha, Seungki Choi, and 4 more authors

   Feb 2021

   Korean Patent
3. US, KR

   Method for detecting chemical contamination clouds and unmanned aerial vehicle performing method

   Hyunwoo Nam, Myung-Kyu Park, Myeongjae Lee, and 2 more authors

   Jul 2021

   US Patent (Jul. 2021), Korean Patent (Dec. 2020)

2020

1. US, KR

   Apparatus and method for estimating location of pollutant source

   Sangwon Do, Eugene Chong, Myeongjae Lee, and 2 more authors

   Oct 2020

   US Patent (Oct. 2020), Korean Patent (Nov. 2020)
2. UK, KR

   Apparatus and method for determining effective signals in digital pulses

   Eugene Chong, Hyunwoo Nam, Young Jin Koh, and 2 more authors

   Oct 2020

   UK Patent (Oct. 2020), Korean Patent (Dec. 2018)