Skip to reading
AI.Foundations.DL AI / Modern ML

Deep learning — backprop, optimization, regularization

Node 5 of 12 in AI

Objectives

  • Identify the key rules and §§ that apply to deep learning — backprop, optimization, regularization.
  • Apply the AI.Foundations.DL knowledge element in a typical exam scenario.
  • Recognize common distractors and partial-credit answers.

Read

AI.Foundations.DL

Deep learning — backprop, optimization, regularization

AI Era · Deep Learning Neural networks, optimizers, regularization

**Backpropagation** computes ∂L/∂θ via reverse-mode automatic differentiation. Modern frameworks (PyTorch 2.5, JAX) build a computation graph and walk it backward, computing gradients in O(forward) time.

**Optimizers:** • **SGD with momentum** — still used for CNN image training; best generalization for computer vision. • **Adam / AdamW** — adaptive per-parameter learning rate via first and second moment estimates. AdamW decouples weight decay from the adaptive term (Loshchilov & Hutter 2019). Default for transformers. • **Lion, Shampoo** — newer second-order-ish optimizers; Shampoo (Gupta et al.) uses Kronecker-factored preconditioners.

**Learning-rate schedules:** warmup (linear for first ~1-10% of steps), cosine decay, inverse square root. For LLM pretraining, warmup + cosine-to-zero is standard.

**Regularization:** • **Dropout** — zero activations with probability p; inverted dropout scales at train time. Less common in modern LLMs (layer/RMS norm + large batches serve similar role). • **Weight decay** (L2) — shrinks parameters. • **Gradient clipping** — cap ‖∇‖₂ (typ. 1.0) for stability. • **Label smoothing** — ε-mix target with uniform. • **Mixup, CutMix** — input/label interpolation, improves CV robustness.

**Mixed precision** (FP16/BF16 with loss scaling, FP8 in H100/H200) halves memory and doubles throughput. FSDP and DeepSpeed ZeRO stage-3 shard params/grads/opt-states across GPUs. Tensor/sequence/pipeline parallelism compose for frontier-scale training.

**Loss landscape:** modern wide networks have many nearly-equivalent low-loss minima connected by linear mode connectivity; explains why SGD variants find generalizing solutions.

Loshchilov & Hutter (2019) Decoupled Weight Decay, ICLR Micikevicius et al. (2018) Mixed Precision Training

Check yourself

3 quick questions — no score kept, just formative feedback.

  1. Q1 · AI.Foundations.DL

    AdamW (Loshchilov & Hutter 2019) improves upon Adam by:

    Answer choices
  2. Q2 · AI.Foundations.Governance

    The NIST AI Risk Management Framework 1.0 defines four core functions. Which is correct?

    Answer choices
  3. Q3 · AI.Foundations.ML

    The bias-variance decomposition of expected squared error says:

    Answer choices

Tutor

Scoped to AI.Foundations.DL .

  1. Ask questions about this passage. Answers cite the specific corpus chunk and regulation. The tutor will never reproduce real exam items.