DOI QR코드

DOI QR Code

누적 가중치 변화의 시각화를 통한 심층 신경망 분석시스템

Deep Neural Network Analysis System by Visualizing Accumulated Weight Changes

  • Taelin Yang (Department of Digital Media, Soongsil University) ;
  • Jinho Park (Department of Digital Media, Soongsil University)
  • 투고 : 2023.06.16
  • 심사 : 2023.07.05
  • 발행 : 2023.07.25

초록

최근 ChatGPT나 자율주행 자동차 등의 인공지능 분야의 급속한 발전으로 인해 인공지능에 대한 관심이 높아졌다. 그러나 아직 인공지능은 학습 과정에서 알 수 없는 요소가 많이 존재하여 모델을 개선하거나 최적화하기 위해서 필요 이상의 시간과 노력을 들여야 하는 경우가 많다. 따라서, 인공지능 모델의 학습 과정에서 가중치 변화를 명확하게 이해하고 해당 변화를 효과적으로 분석할 수 있는 도구 또는 방법론이 절실하게 요구되고 있다. 본 연구에서는 이러한 점을 해결하기 위해 누적 가중치 변화량을 시각화해주는 시스템을 제안한다. 시스템은 학습의 일정한 기간마다 가중치를 구하고 가중치의 변화를 누적시켜서 누적 가중치로 저장하여 3차원 공간상에 나타내게 된다. 이로 인해 보는 이로 하여금 한눈에 레이어의 구조와 현재의 가중치 변화량이 이해되기 쉽게 구성하였다. 이러한 연구를 통해 인공지능 모델의 학습 과정이 어떻게 진행되는지에 대한 이해와 모델의 성능 향상에 도움이 되는 방향으로 하이퍼 파라미터를 변경할 수 있는 지표를 얻게 되는 등 인공지능 학습 과정의 다양한 측면을 탐구할 수 있을 것이다. 이러한 시도를 통해 아직 미지의 영역으로 여겨지는 인공지능 학습 과정의 일부를 보다 효과적으로 탐색하고 인공지능 모델의 발전과 적용에 기여할 수 있을 것으로 기대된다.

Recently, interest in artificial intelligence has increased due to the development of artificial intelligence fields such as ChatGPT and self-driving cars. However, there are still many unknown elements in training process of artificial intelligence, so that optimizing the model requires more time and effort than it needs. Therefore, there is a need for a tool or methodology that can analyze the weight changes during the training process of artificial intelligence and help out understatnding those changes. In this research, I propose a visualization system which helps people to understand the accumulated weight changes. The system calculates the weights for each training period to accumulates weight changes and stores accumulated weight changes to plot them in 3D space. This research will allow us to explore different aspect of artificial intelligence learning process, such as understanding how the model get trained and providing us an indicator on which hyperparameters should be changed for better performance. These attempts are expected to explore better in artificial intelligence learning process that is still considered as unknown and contribute to the development and application of artificial intelligence models.

키워드

참고문헌

  1. Tjoa, Erico, and Cuntai Guan. "A survey on explainable artificial intelligence (xai): Toward medical xai." IEEE transactions on neural networks and learning systems 32.11 (2020): 4793-4813. https://doi.org/10.1109/TNNLS.2020.3027314
  2. Arrieta, Alejandro Barredo, et al. "Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI." Information fusion 58 (2020): 82-115. https://doi.org/10.1016/j.inffus.2019.12.012
  3. Van der Velden, Bas HM, et al. "Explainable artificial intelligence (XAI) in deep learning-based medical image analysis." Medical Image Analysis (2022): 102470.
  4. Zhou, Bolei, et al. "Learning deep features for discriminative localization." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
  5. Ribeiro, Marco Tulio, Sameer Singh, and Carlos Guestrin. "" Why should i trust you?" Explaining the predictions of any classifier." Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. 2016.
  6. Kumar, Siddharth Krishna. "On weight initialization in deep neural networks." arXiv preprint arXiv:1704.08863 (2017).
  7. Bergstra, James, et al. "Algorithms for hyper-parameter optimization." Advances in neural information processing systems 24 (2011).
  8. Liashchynskyi, Petro, and Pavlo Liashchynskyi. "Grid search, random search, genetic algorithm: a big comparison for NAS." arXiv preprint arXiv:1912.06059 (2019).
  9. Frazier, Peter I. "A tutorial on Bayesian optimization." arXiv preprint arXiv:1807.02811 (2018).
  10. Simonyan, Karen, Andrea Vedaldi, and Andrew Zisserman. "Deep inside convolutional networks: Visualising image classification models and saliency maps." arXiv preprint arXiv:1312.6034 (2013).