I have multiple objectives, such as $f(\mathbf{x})$, $g(\mathbf{x})$, and $h(\mathbf{x})$.

I would like to find a set of $x$ that can $\underset{\mathbf{x}}{argmin} [ f(\mathbf{x}) + g(\mathbf{x}) + h(\mathbf{x}) ]$.

However, the three objectives have quite different scales. For instance the output of $f(\mathbf{x})$ is normally in [1000, 5000], while that of $g(\mathbf{x})$ is in [0, 1].

Does this negatively impact the oprimization? To compute the $argmin$, should I normalize them to an identical scale?

  • $\begingroup$ It depends on what the various objectives mean. When there's no meaningful interpretation, it's common to add tunable parameters as coefficients: $f(\mathbf{x}) + \lambda g(\mathbf{x}) + \gamma h(\mathbf{x})$. By changing $\lambda$ and $\gamma$, you change the relative importance of $f$, $g$, and $h$ in the overall optimization. $\endgroup$ – tddevlin Apr 11 at 17:58

As mentioned in a comment by @tddevlin, it depends on what those objectives are. I give you an example. Suppose I'm solving a vehicle routing problem and I want to minimize both cost and number of vehicles. Then, my $f(x)$ is cost and $g(x)$ is the number of trucks, which both fit the profile that you mentioned. Now, if I prefer costs over the number of trucks, I don't even need to add any weight to my $f$ and $g$ functions. In fact, I can use them exactly as is and know that because of the small scale of $g$ compared to $f$, always $f$ will be preferred first, which is the desired behavior. That's why it matters what are your objectives.


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.