Deep Learning Foundations

Tracks

1. 01🧠Why Deep Learning?

   0/10 lessons

   Representations, scale, and when not to use it

   Deep learning is useful when representation learning, differentiable optimization, and enough data or pretraining beat hand-crafted features.

   Lesson list (10)

   • 01
     Hand-Crafted Features Hit Limits~22 min · features, representation, history
   • 02
     Classical ML vs Deep Learning~22 min · classical-ml, tradeoffs, baselines
   • 03
     Representation Learning~22 min · representation, embeddings, transfer
   • 04
     The History Arc~18 min · history, alexnet, transformer
   • 05
     Data, Compute, and Hardware~18 min · gpu, tpu, scaling
   • 06
     What Neural Networks Excel At~16 min · use-cases, applications
   • 07
     Where Deep Learning Is Overkill~16 min · anti-patterns, tradeoffs
   • 08
     The Cost of Deep Learning~16 min · cost, infra, ops
   • 09
     The Modern Stack~16 min · pytorch, huggingface, ecosystem
   • 10
     Roadmap: What You Will Learn~12 min · roadmap, tracks

   ✓Quiz · 4 questions→

2. 02🧩Neurons, Layers, and Forward Passes

   0/8 lessons

   From linear maps to output heads

   Networks are composed transformations with explicit shape contracts.

   Lesson list (8)

   • 01
     The Artificial Neuron~22 min · neuron, linear, activation
   • 02
     Perceptron Basics~18 min · perceptron, history
   • 03
     Why a Single Linear Layer Is Limited~18 min · linearity, xor, depth
   • 04
     Multilayer Perceptrons~25 min · mlp, fully-connected, depth
   • 05
     Activation Functions~22 min · relu, gelu, softmax
   • 06
     Width, Depth, and Expressive Power~18 min · capacity, depth, scaling
   • 07
     The Forward Pass~22 min · forward, shapes, autograd
   • 08
     Batches, Logits, and Output Heads~22 min · batches, logits, heads

   ✓Quiz · 4 questions→

3. 03⛓️Loss, Gradients, and Backpropagation

   0/9 lessons

   How neural networks actually learn

   Learning means reducing loss through gradients, optimizer state, and stable update rules.

   Lesson list (9)

   • 01
     What It Means to Learn~18 min · learning, loss, gradient-descent
   • 02
     Loss Functions~22 min · loss, cross-entropy, mse
   • 03
     Gradient Intuition~18 min · gradients, calculus, geometry
   • 04
     Gradient Descent and Mini-Batches~22 min · sgd, minibatch, noise
   • 05
     Backpropagation: The Chain Rule~25 min · backprop, chain-rule, autograd
   • 06
     Vanishing and Exploding Gradients~22 min · vanishing, exploding, depth
   • 07
     Optimizers: SGD to AdamW~22 min · optimizer, adamw, momentum
   • 08
     Learning Rate Schedules~22 min · lr, schedule, warmup
   • 09
     The Training Loop~25 min · training-loop, epochs, validation

   ✓Quiz · 4 questions→

4. 04🏋️Training Data, State, and Checkpoints

   0/8 lessons

   Dataloaders, modes, saving, and evaluation

   Training is a stateful workflow, not one heroic code cell.

   Lesson list (8)

   • 01
     Dataset Objects~18 min · dataset, pytorch, data
   • 02
     DataLoader and Batching~20 min · dataloader, batching, workers
   • 03
     Devices and Tensor Placement~18 min · device, cuda, mps
   • 04
     One Train Step~18 min · train-step, zero-grad, step
   • 05
     Evaluation Mode~16 min · eval, dropout, batchnorm
   • 06
     Metrics During Training~18 min · metrics, logging, monitoring
   • 07
     Checkpointing~18 min · checkpoint, save, resume
   • 08
     Reproducibility and Seeds~16 min · seed, reproducibility, determinism

   ✓Quiz · 4 questions→

5. 05🛡️Stability, Regularization, and Making Training Work

   0/10 lessons

   Overfitting, curves, normalization, residuals

   Most deep-learning work is making training stable and honest.

   Lesson list (10)

   • 01
     Overfitting in Neural Networks~18 min · overfitting, generalization, capacity
   • 02
     Reading the Loss Curves~18 min · debugging, loss-curves
   • 03
     Dropout~18 min · dropout, regularization
   • 04
     Weight Decay and Early Stopping~18 min · weight-decay, early-stopping, l2
   • 05
     BatchNorm and LayerNorm~22 min · batchnorm, layernorm, normalization
   • 06
     Initialization: Xavier and He~18 min · init, xavier, he
   • 07
     Residual Connections~22 min · residual, skip, resnet
   • 08
     Gradient Clipping~14 min · clipping, stability
   • 09
     Data Augmentation~18 min · augmentation, regularization
   • 10
     Mixed Precision and Reproducibility~18 min · amp, fp16, bf16

   ✓Quiz · 4 questions→

6. 06🏗️Architectures: Vision, Sequence, and Attention

   0/12 lessons

   CNNs, recurrence, attention, transformers

   Architectures encode assumptions about the structure of data.

   Lesson list (12)

   • 01
     Why Images Are Different~16 min · images, spatial, translation-invariance
   • 02
     Convolutions: Shared Filters~22 min · conv, filters, stride
   • 03
     CNN Intuition: Edges to Objects~18 min · cnn, feature-hierarchy, interpretability
   • 04
     Classic CNN Lineage~18 min · cnn, history, resnet
   • 05
     Transfer Learning in Vision~18 min · transfer-learning, fine-tuning, vision
   • 06
     Vision Transformers~18 min · vit, transformer, patches
   • 07
     Practical Vision Tasks~16 min · detection, segmentation, classification
   • 08
     Why Sequence Data Is Hard~16 min · sequence, rnn, context
   • 09
     Recurrent Neural Networks~18 min · rnn, lstm, gru
   • 10
     LSTMs and GRUs~18 min · lstm, gru, gating
   • 11
     The Attention Breakthrough~22 min · attention, self-attention, transformer
   • 12
     Transformer Blocks~22 min · transformer, blocks, ffn

   ✓Quiz · 4 questions→

7. 07🧪Modern Deep Learning Practice

   0/8 lessons

   PyTorch mental model, transfer, accelerators, tracking

   Practice is where deep learning becomes a repeatable workflow instead of a notebook stunt.

   Lesson list (8)

   • 01
     PyTorch Mental Model~22 min · pytorch, mental-model, modules
   • 02
     Training Loop Anatomy~18 min · training-loop, scaffold
   • 03
     Transfer Learning vs Fine-Tuning~16 min · transfer, fine-tune, freezing
   • 04
     Parameter-Efficient Tuning: LoRA~22 min · lora, peft, fine-tune
   • 05
     Accelerators and Devices~16 min · gpu, tpu, mps
   • 06
     Mixed Precision~16 min · amp, bf16, fp16
   • 07
     Experiment Tracking~18 min · tracking, wandb, mlflow
   • 08
     Compilation and Export Paths~18 min · onnx, torchscript, compile

   ✓Quiz · 4 questions→

8. 08🗺️Choosing the Right Approach

   0/8 lessons

   Scale, foundation models, safety, and final map

   The final skill is knowing when deep learning is the right tool and what kind of deep learning to use.

   Lesson list (8)

   • 01
     When Deep Learning Beats Classical ML~16 min · choice, tradeoffs, classical-ml
   • 02
     Choosing: MLP, CNN, Transformer, Foundation Model~18 min · choice, architecture
   • 03
     Parameter Count and Scaling~16 min · scaling, parameter-count, compute
   • 04
     Foundation Models~18 min · foundation-model, llm, transfer
   • 05
     Fine-Tuning vs Prompting vs Feature Extraction~16 min · adaptation, tradeoffs
   • 06
     Robustness, Bias, and Safety~18 min · safety, bias, robustness
   • 07
     Staying Current Without Drowning~16 min · learning, research, habits
   • 08
     Final Map: From Basics to Modern AI~22 min · recap, roadmap

   ✓Quiz · 5 questions→