I understanding how gradient is calculated in the usual context--it is just taking partial derivative w.r.t. each element of the X vector. Say a function $f$ has $n$ independent variables, denoted by $x_1$ ... $x_n$, the gradient of $f$ is: $$ \nabla f = \left<\frac{\partial f}{\partial x_1};\frac{\partial f}{\partial x_2};...;\frac{\partial f}{\partial x_n}\right> $$

However, when it comes to the gradient of the loss function of a neural network, a new term, backpropagation, emerges. My confusion is I am not sure if backpropagation makes the above calculation different or is it just another name of it.

Let me use the following simple neural network as an example (courtesy of this very informative video):

enter image description here

This NN has two nodes, four weights (denoted by $w_1$ ... $w_4$) and three biases (denoted by $b_1$ ... $b_3$). So the loss function is defined like this:

$L(w_1, w_2, w_3, w_4, b_1, b_2, b_3)$ = [a loss function, which can be extremely lengthy but differentiable]

Without knowing the exact meaning of backpropagation, I would say its gradient is just: $$ \nabla L = \left<\frac{\partial L}{\partial w_1};\frac{\partial L}{\partial w_2};...;\frac{\partial L}{\partial b_3}\right> $$

and the calculation of the partial derivative w.r.t. $w_1$, $w_2$ ... $b_2$, $b_3$ is no more than repetitive application of chain rules, etc--we don't need anything more than good old calculus to make it work. With gradient calculated, we can use gradient descent to minimize the loss function (well, the convexity/local minima is another big issue, let's ignore it for the purpose of this particular question).

So it begs the question: what is the role of "backpropagation" in the calculation? Does backpropagation make the calculation easier (i.e., backpropagation is faster but the resultant gradient will be exactly the same as the traditional partial derivative approach) or is it just a name of this calculation?

  • 2
    $\begingroup$ In a sentence: Yes, it is "just" the chain rule, but it's in fact, NP complete to figure out the most efficient order in which to apply the chain rule, it's not always trivial to figure out how to multiply everything together. $\endgroup$
    – shimao
    Dec 4, 2021 at 5:11
  • $\begingroup$ Hi @shimao, I get the first half of your comment "just" the chain rule but I am not quite sure what exactly you are trying to say about the "NP complete" part. Do you mean that backpropagation is NOT a deterministic algorithm (such as quick sort or Dijkstra's shortest paths) which has a fixed time complexity? In other words, do you mean backpropagation is more similar to an algorithm "solving" travelling salesman problem--it only gets an "okay" result but an optimal result is NOT guaranteed? $\endgroup$
    – Alex Kong
    Dec 4, 2021 at 5:47
  • $\begingroup$ You have to think computationally.Any differentiation on a neural network will use the chain rule. you could calculate the gradient of each weight independently (without reusing intermediate results). Back propagation is the calculation by first finding errror derivative with respect to output layer, then using that to calculate gradient wrt weights leading into output layer... So its a particular way to efficiently structure your gradient calculations for a NN. But I would agree its pretty obvious and used in most other areas of nonlinear optimisation. $\endgroup$
    – seanv507
    Dec 4, 2021 at 7:11
  • $\begingroup$ @shimao is saying that the choice of how to structure your chain rule calculations is NP complete. so any particular backpropagation algorithm is deterministic, finding the most efficient is NP complete $\endgroup$
    – seanv507
    Dec 4, 2021 at 7:25
  • $\begingroup$ @seanv507 You have to think computationally. -> so you mean back-propagation is faster? you could calculate ..(without reusing intermediate results). Back propagation ... first finding errror derivative with respect to output layer, then using that to calculate gradient wrt weights leading into output layer... -> My understanding is that, back-propagation is similar to recursion and it reuses some intermediate results so that these results are't calculated time and time again. Is this understanding correct? $\endgroup$
    – Alex Kong
    Dec 4, 2021 at 9:22

1 Answer 1


Here are some examples to illustrate how backpropagation (automatic differentiation) is quite nontrivial, and the tradeoffs involved.

Consider $x_n = f_n(x_{n-1}) = f_n(f_{n-1}(x_{n-2})) = f_n(f_{n-1}(\ldots f_1(x_0)))$. Let $J_i(x)$ denote the jacobian of $f_i$ at $x$. Then it's true that $\frac{d x_n}{d x_0} = J_n(x_{i-1}) J_{n-1}(x_{i-1}) \ldots J_1(x_0)$.

But what order should you do the multiplication in? For example, if $n = 3$, and the dimension of $x_0, x_1, x_2, x_3$ are 1000, 100, 10, 1 respectively, then it makes sense to multiply from left to right, so the most expensive matrix multiplication you'll do is a $1 \times 100$ by $100 \times 1000$. If you go right to left, you'll end up doing a $10 \times 100$ by $100 \times 1000$ matmul, which is 10 times slower. The former method is called "reverse mode", and the latter is called "forward mode". Backpropagation is typically implemented using reverse mode, because it is more suited to neural network training (and in general, when the number of inputs is greater than the number of outputs).

Now consider if the dimension of $x_0, x_1, x_2, x_3, x_4$ are 1000, 1, 1000, 1, 1000. Both forward and reverse mode accumulation will result in a $1000 \times 1000$ by $1000 \times 1$ product, whereas if you multiply the two middle jacobians first, you only end up doing $1000 \times 1$ by $1 \times 1$. So this alternate multiplication order is much more efficient. Ironically, there is no efficient algorithm for determining the most efficient order - this is an NP-complete problem.

Another consideration is that typical jacobians are extremely sparse: Applying pointwise sigmoid activation to an vector of size 1 million creates a $1000000 \times 1000000$ jacobian matrix, with only 1 million non-zero values. It's infeasible to multiply jacobians this large, but to carry out backpropagation, it's actually only necessary to compute the "vector jacobian product": $\text{vjp}_i(v,x) = v^TJ_i(x)$. Then if $x_n$ has dimension 1 (i.e any loss function), $\frac{d x_n}{d x_0} = \text{vjp}_1( \text{vjp}_2( \ldots\text{vjp}_n(1,x_{n-1})\ldots, x_1), x_0)$ carries out reverse mode backprop. For a pointwise function $g$, $\text{vjp}_g(v,x) = v\ \circ \ g'(x)$ is trivial to compute.

So now suppose we've implemented blazing fast reverse mode automatic differentiation with vjps, and we want to minimize a function using second order information from the hessian matrix. We could compute the hessian by first computing the gradient $\nabla(x)$, and then backpropagate the gradient calculation itself: $H_i = \text{vjp}_{\nabla_i}(1, x)$.

Note that because we're using vjps / reverse mode backprop, we can only compute one row of the hessian at a time - as noted above, reverse mode is poorly suited when the dimension of the output is large. Ideally, we'd want to use reverse mode for computing the gradient, but forward mode to differentiate the gradient. As it turns out, you can implement a "jacobian vector product": $\text{jvp}_i(x,v) = J_i(x)v$ as two invocations of vjp (link below). A vjp is to reverse mode as jvp is to forward mode, and this enables us to efficiently perform second order optimization.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.