dongwook_
  • WorkWork
  • LabLab
  • LogLog▾
    • Computer vision8
      • Classification4
      • Detection2
      • Segmentation2
    • Deep learning2
    • Machine learning2
    • Statistics4
  • AboutAbout

log

Deep learning

  • 2026-09-15Deep learning

    제약이 능력을 만든다: PyTorch로 이해하는 AE, DAE, VAEConstraints Create Capability: Understanding AE, DAE and VAE with PyTorch

    입력을 그대로 출력하는 걸 왜 학습할까? 병목·손상된 입력·확률분포라는 제약에 따라 AE, DAE, VAE가 무엇을 배우는지 MNIST로 실험한다.Why train a model to output its own input? Bottlenecks, corrupted inputs and probability distributions on MNIST — what AE, DAE and VAE each learn.

    • #AE
    • #AutoEncoder
    • #DAE
    • #DL
    • #DeepLearning
    • #Denoising AutoEncoder
    • #PyTorch
    • #VAE
    • #Varitional AutoEncoder
    • #능력
    • #제약
  • 2026-09-07Deep learning

    Optimizer는 무엇을 보완해 왔는가?What Has Each Optimizer Been Fixing?

    Gradient Descent부터 Momentum, AdaGrad, RMSProp, Adadelta, Adam까지 — 각 Optimizer가 어떤 상태값을 저장하는지를 축으로 PyTorch로 직접 구현해 본다.From plain gradient descent to Momentum, AdaGrad, RMSProp, Adadelta and Adam — implemented by hand in PyTorch, framed by what state each one stores.

    • #AdaGrad
    • #DL
    • #DeepLearning
    • #Gradient Descent
    • #Mini batch
    • #Optimizer
    • #RMSProp
    • #SGD
    • #adam
    • #momentum
    • #경사하강법
    • #옵티마이저
    • #확률적 경사 하강법

© 2026 dongwook · Built with Astro