You can have the journal publish an erratum, if the changes are limited and bear on statements of fact. For example, if you inverted the array indices in proof 3, or a typo in lemma 2 changes its meaning, or you quoted from another paper and forgot the citation. The journal's web site will show a link to the update.
Otherwise, the published version is the version of record, and wholesale changes wouldn't normally be acceptable for two reasons. First, the journal would have to ask the reviewers (in this case, ad hoc reviewers for a conference with which the journal is not associated) to re-evaluate the manuscript. Second, other scientists may have already based work on the published version in good faith.
(An analogy, if you are a user of the
git version control system, is forcing a push to a public repository. Don't be that person.)
If you can substantially improve the original by these changes, then it is worthwhile to make an updated version available through ArXiv or on an institutional archive. At least it will be indexed and will show up in searches. It is common to see this kind of update in CVs and publication lists.
If the changes are only cosmetic, or bear on things like citing existing work or improving figures, then my suggestion is to stick with the old version. Everybody has papers that they would have liked to improve, but couldn't. Save your energy for the next paper!