Calculus

44 problems

• Adadelta optimizerMedium
• Adagrad optimizerEasy
• Adam optimiser stepMedium
• Adamax optimizerMedium
• AdamW step (decoupled weight decay)Medium
• Basic autograd operationsMedium
• Conv2D backward (dx, dW)Hard
• ReLU forward + backwardEasy
• Sigmoid forward + backwardEasy
• MSE loss + backwardEasy
• Linear backward (chain rule)Medium
• SGD step (in place)Easy
• Clipped Inverse Propensity WeightingMedium
• Warmup + cosine decay LR scheduleEasy
• Cross-entropy gradientMedium
• Backprop — all four gradients for one hidden layerMedium
• Backprop — the output deltaEasy
• Perceptron — the learning ruleMedium
• Sigmoid derivative from the activationEasy
• Vanishing gradients through sigmoid layersEasy
• Backprop — the weight gradient is an outer productEasy
• Dropout layer (forward & backward)Medium
• Elastic-Net regression (gradient descent)Medium
• Exponential LR schedulerEasy
• GELU backward (tanh approximation)Medium
• The residual Jacobian I + ∂f/∂xEasy
• Clip gradients by global L2 normEasy
• Learning-rate range testMedium
• Linear regression — gradient descentEasy
• Linear regression — one gradient descent stepEasy
• Lion optimizer stepMedium
• Logistic regression — gradient descentMedium
• Logistic regression — the BCE gradient (error × feature)Easy
• Matmul backwardMedium
• Nesterov accelerated gradientMedium
• Neural ODE forward EulerMedium
• RMSprop optimizerEasy
• BPTT — the gradient-norm trajectoryMedium
• Vanishing or exploding? Classify the regimeEasy
• BPTT — the tanh backward factorEasy
• SGD with momentum (one step)Easy
• Softmax backwardMedium
• Softmax (multinomial) regressionMedium
• Step LR schedulerEasy