Note_Tech

All technological notes.

---

Project maintained by simonangel-fong Hosted on GitHub Pages — Theme by mattgraham

Deep Learning

Back

---

• Deep Learning

• Environment Configuration

---

• Neural Network

  • Linear Neural Networks for Classification
  • Multilayer Perceptrons (MLP)

---

• Summary:

• Essential part of DL:

  • data:
    • tensor

  • model:

    • forward function
    • activation function
      • linear: ax
      • simoid: (0, 1), output layer with two categories.
      • relu: (0, x), hidden layer
      • softmax: %, output layers with more than two categories

  • Objective function:

    • loss fnction
    • criterion
      • MSELoss: linear regression
      • CrossEntropyLoss: multi-class classification
  • Algorithm:
    • optimizer
      • Backward Propagation: find gradient
      • gradient descent:
        • find gradient + update weights
        • lr (how fast model learns) + gradient (direction the model learns)
        • rely Backward Propagation
        • optim.SGD(params)
      • optim.Adam(params)

---

TOP