Chain rule shown with the explicit sum: \(\frac{\partial e}{\partial a}=\sum_i \frac{\partial e}{\partial t_i}\frac{\partial t_i}{\partial a}\)
e = (y − (w x + b))² + w²
Output and input gradients
e = —
∂e/∂w=— ∂e/∂x=— ∂e/∂b=—
Selected node (click a node in the graph)
—
Entire calculation for the selected node
Click a node to view its full chain-rule expansion.
Entire calculation for the current backward node
None yet.
In this graph, a variable like \(w\) influences \(e\) through multiple children (\(t_1\) and \(t_5\)).
The demo always shows the sum over children explicitly, matching the notebook notation.
Computation graph
Nodes show computation, value, and \(\frac{\partial e}{\partial v}\)
phase: idle
forward computed
derivative computed
active
Values and derivatives
Derivatives written as \(\frac{\partial e}{\partial v}\)