Deep Learning Terms Every Beginner Should Understand

Core Terms — Plain English:
• Neural Network → A model made of layers of neurons
• Layer → One step in the model's data flow
• Neuron → A small unit that does math on inputs
• Weight → How strongly one neuron influences another
• Activation → A function that adds non-linearity
• Forward Pass → Data flows from input to output

Where You'll See Them:
☐ In model definition code (nn.Module)
☐ In architecture diagrams
☐ In paper abstracts
☐ In interview questions
☐ In debugging stack traces
☐ In training loop logs

Training Terms — Plain English:
• Loss → "How wrong am I?"
• Optimizer → "How should I update my weights?"
• Backpropagation → "How do I compute the gradients?"
• Gradient → "Which direction is downhill?"
• Epoch → "I've seen all my data once"
• Batch Size → "How many samples per step"
• Learning Rate → "How big is each step"

Typical Training Loop:
☐ for epoch in range(num_epochs):
☐   for batch in dataloader:
☐     forward pass → predictions
☐     loss = criterion(predictions, targets)
☐     loss.backward()  ← backpropagation
☐     optimizer.step() ← update weights
☐     optimizer.zero_grad()

Evaluation Terms — Plain English:
• Overfitting → Great on train, bad on test
• Underfitting → Bad on both train and test
• Bias → Wrong assumptions (too simple)
• Variance → Too sensitive to data noise
• Validation Set → Where you tune
• Test Set → Where you judge
• Regularization → Controls overfitting

Bias-Variance Quick Guide:
High Bias + Low Variance  → Underfitting
Low Bias + High Variance  → Overfitting
Low Bias + Low Variance   → The goal
High Bias + High Variance → Model is broken

Signals:
□ Train loss high + test loss high → underfitting
□ Train loss low + test loss high → overfitting
□ Both low → good fit

Term Mastery Tracker:
☐ Neural Network — can explain in 1 sentence
☐ Layer — can point to one in code
☐ Neuron — can describe weighted sum
☐ Weight — can explain what is learned
☐ Activation Function — know 2 examples
☐ Forward Pass — can trace data flow
☐ Loss — can name 3 loss functions
☐ Optimizer — can name 2 optimizers
☐ Backpropagation — can explain gradient flow
☐ Epoch — can interpret training logs
☐ Batch Size — can explain trade-offs
☐ Learning Rate — can explain too high/too low
☐ Overfitting — can spot in a loss curve
☐ Underfitting — can spot in a loss curve
☐ Bias / Variance — can explain trade-off
☐ Validation Set — can describe its purpose

Weekly Review — Term Check:
Week: _____

Terms I fully understand:
1. 
2. 
3. 

Terms I'm still shaky on:
1. 
2. 
3. 

Terms I used in code this week:
1. 
2. 
3. 

Next week's focus terms:
1. 
2. 
3. 

Concept Map — How Terms Connect:
Input
  ↓
Layer (neurons + weights + activations)
  ↓
Forward Pass → Prediction
  ↓
Loss (how wrong?)
  ↓
Backpropagation (gradients)
  ↓
Optimizer (update weights)
  ↓
Repeat over epochs
  ↓
Evaluate on validation + test set
  ↓
Check for overfitting / underfitting

If you can draw this, you understand the terms.