Deep Learning problems
Activations, forward pass, backprop. The core of every neural net.
455 problems166 Easy202 Medium87 Hard
50 problems
ReLU
dl-001
Sigmoid (Vector)
dl-002
Numerically Stable Softmax
dl-003
MLP Forward Pass
dl-004
Backprop Gradient
dl-005
Leaky ReLU
dl-006
ELU
dl-007
Tanh (Vector)
dl-008
One-Hot Encode
dl-009
Dense Layer Forward
dl-010
Conv Output Dimension
dl-011
ReLU Backward
dl-012
Sigmoid Backward
dl-013
Tanh Backward
dl-014
Causal Mask
dl-015
SGD Weight Update
dl-016
Warmup LR
dl-017
Xavier Init Std
dl-018
He Init Std
dl-019
Embedding Lookup
dl-020
Temperature Softmax
dl-021
Conv2D Single Channel Forward
dl-022
SAME Padding Sizes
dl-023
Max Pooling Forward
dl-024
BatchNorm Forward (Train)
dl-025
BatchNorm Inference
dl-026
LayerNorm Forward
dl-027
Inverted Dropout (Seeded)
dl-028
Momentum Update
dl-029
Cosine Annealing LR
dl-030
Linear Layer Backward dW
dl-031
Embedding Backward Scatter-Add
dl-032
Sinusoidal Positional Encoding
dl-033
Scaled Attention Scores
dl-034
Attention Softmax Weights
dl-035
Attention Weighted Sum
dl-036
Multi-Head Split
dl-037
Multi-Head Merge
dl-038
RNN Forward One Step
dl-039
Gradient Clipping (Global Norm)
dl-040
Perplexity From Logits
dl-041
GELU (Tanh Approximation)
dl-042
Softmax + Cross-Entropy Backward
dl-043
BatchNorm Backward (Simplified)
dl-044
LSTM Gates Forward
dl-045
GRU Gates Forward
dl-046
Beam Search One Step (Beam 2)
dl-047
Label Smoothing Cross-Entropy
dl-048
Numeric Gradient Check (L2 Error)
dl-049
MLP FLOPs
dl-050
Showing 50 of 455 problems.