Efficient AI

Efficient neural modeling of visual information

Designing neural network architectures is one of the most fundamental research topics in the field of AI. It involves crafting the appropriate structure and operation of AI to effectively and efficiently model specific problems and data. For visual information, convolutional neural network is the most popular architecture. It is specialized for learning local information, but has limitations in modeling global information. While simply stacking more layers can achieve desirable performance by increasing receptive field of neural networks, it results in unnecessarily large networks. Therefore, we have designed neural architectures with attention mechanisms, focusing on not only fully utilizing information (i.e., both local and global information) in images but also optimizing efficiency in terms of both the number of model parameters and computational complexity (Kim & Lee, 2018; Kim et al., 2020; Kim et al., 2022; Kim et al., 2024). We have also designed efficient neural networks with recursive or multi-exit structures (Choi et al., 2021; Jeon et al., 2020). In addition, we have explored network pruning for image compression models (Kim et al., 2020).

References

2024

  1. NeurIPS
    Diversify, contextualize, and adapt: Efficient entropy modeling for neural image codec
    Jun-Hyuk Kim, Seungeon Kim, Won-Hee Lee, and Dokwan Oh
    Advances in Neural Information Processing Systems, 2024
    arXiv HTML

2022

  1. CVPR
    Joint global and local hierarchical priors for learned image compression
    Jun-Hyuk Kim, Byeongho Heo, and Jong-Seok Lee
    IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022
    arXiv HTML

2021

  1. Volatile-nonvolatile memory network for progressive image super-resolution
    Jun-Ho Choi, Jun-Hyuk Kim, Manri Cheon, and Jong-Seok Lee
    IEEE Access, 2021
    HTML Code

2020

  1. MAMNet: Multi-path adaptive modulation network for image super-resolution
    Jun-Hyuk Kim, Jun-Ho Choi, Manri Cheon, and Jong-Seok Lee
    Neurocomputing, 2020
    arXiv HTML Code
  2. LarvaNet: Hierarchical super-resolution via multi-exit architecture
    Geun-Woo Jeon, Jun-Ho Choi, Jun-Hyuk Kim, and Jong-Seok Lee
    European Conference on Computer Vision Workshop, 2020
    HTML
  3. Efficient deep learning-based lossy image compression via asymmetric autoencoder and pruning
    Jun-Hyuk Kim, Jun-Ho Choi, Jaehyuk Chang, and Jong-Seok Lee
    IEEE International Conference on Acoustics, Speech and Signal Processing, 2020
    HTML

2018

  1. Deep residual network with enhanced upscaling module for super-resolution
    Jun-Hyuk Kim and Jong-Seok Lee
    IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2018
    HTML Code