My research interest focuses on redesigning systems with an AI approach.
My research interest focuses on redesigning systems with an AI approach.
Hwijoon Lim, Wei Bai, Yibo Zhu, Youngmok Jung and Dongsu Han
Hyunho Yeo, Chan Ju Chong, Youngmok Jung, Juncheol Ye and Dongsu Han
Jaehong Kim, (CO-First)Youngmok Jung, Hyunho Yeo, Juncheol Ye and Dongsu Han
Hyunho Yeo, Youngmok Jung, Jaehong Kim, Jinwoo Shin, and Dongsu Han
Redesigning short-read alignment software with a machine-learning approach
- Developed a short read alignment tool that is the first to leverage learned indices to find exact match of
short substring (0-300 length) in the long reference DNA (6 billion length for human genome)
- Designed new Learned-index structure and replaced the traditional indexing methods used in alignment software
- Achieved up to 3.45x speedup in throughput over state-of-the-art algorithm by reducing the number
of instructions by 4.60x, memory accesses by 8.77x, and LLC misses by 2.21x, while ensuring
identical output.
Redesigning Live-video delivery system with a Deep-learning approach (SIGCOMM 2020)
- LiveNAS system applies deep-learning based super-resolution on the live-video stream to enhance the original live-video stream
- Implemented new APIs in C++ WebRTC open-source code from Google; 1) to obtain decoded and
encoded frames with the frame timestamp. 2) to control and obtain bandwidth information in the
WebRTC.
- Developed the LiveNAS system which involves; 1) Online training and inferencing super-resolution
DNN model during the live video stream. 2) Transmitting training data over RTP. 3) Controlling bandwidth allocation
- LiveNAS system delivers live video with the same quality as WebRTC using only 45.9% bandwidth on average or enhances average QoE by 69% compared to WebRTC using the same bandwidth.
Timeout-less Transport Protocol for Datacenter (EuroSys 2021)
- Implemented NS-3 simulation code[C++] for RDMA protocol and TLT.
- Managed switch configuration which involves priority-flow-control, Shared Buffer Pool size, and Packet
Color.
- TLT protocol reduces 99%-ile FCT up to 91.7% on handling bursts of SET operations
Deep Learning-based Internet Video Delivery (OSDI 2018)
- Developed an integrated-ABR (adaptive bit-rate) controller code[Python,Tensorflow] based on Reinforcement Learning (actor-critic framework of A3C).
- Developed simulation code for video delivery to train RL model and measure user QoE (Quality of Experience).
- Implemented evaluation code to run 5,832 hours of real-time video streaming in Google Cloud in just a few days.
- NAS system enhances the average QoE by 43.08% using the same bandwidth budget or saves 17.13% of bandwidth while providing the same user QoE.