Suppose you have a formula $$ \sum_{n\geq 0}f(n)\frac{x^n}{n!}=\exp\left(x+\frac{x^2}{2}\right). $$
There is a recurrence for $f(n)$ found by differentiation, $$ \sum_{n\geq 1}f(n)\frac{x^{n-1}}{(n-1)!}=(1+x)e^{x+x^2/2}=(1+x)\sum_{n\geq 0}f(n)\frac{x^n}{n!}. $$ So equating coefficients gives $f(n+1)=f(n)+nf(n-1)$ for $n\geq 1$.
Moreover, and explicit formula is found by noting $$ \sum_{n\geq 0}f(n)\frac{x^n}{n!}=e^xe^{x^2/2}=\left(\sum_{n\geq 0}\frac{x^n}{n!}\right)\left(\sum_{n\geq 0}\frac{x^{2n}}{2^nn!}\right) $$ and so $$ f(n)=\sum_{i\geq 0\atop\text{$i$ even}}\binom{n}{i}\frac{i!}{2^{i/2}(i/2)!}=\sum_{j\geq 0}\binom{n}{2j}\frac{(2j)!}{2^jj!}. $$
Based on this, what is a nice expression for $\sum_{i=0}^n(-1)^{n-i}\binom{n}{i}f(i)$? (I am grateful for Sasha's current answer, but is it possible to derive such an expression more combinatorially without reference to random variables and moments? If not, that is no problem. I'm just glad to see and answer.)
Thanks!
(I should note that this is motivated by a passage in Richard Stanley's Enumerative Combinatorics following example 1.1.15.)
The function $\mathcal{M}(t) = \exp(t+t^2/2)$ is the moment generating function for the normal random variable $Z$ with mean $\mu =1$ and variance $\sigma^2 = 1$. This makes $f(n) = \mathbb{E}(Z^n)$ moments of $Z$.
The expression (asumming $(-1)^{n-1}$ was meant to be $(-1)^{n-r}$): $$ g(n) = \sum_{r=0}^n (-1)^{n-r} \binom{n}{r} f(r) = \mathbb{E}((Z-1)^n) $$ which is a multiple of the central moment of $Z$. But these are well know to be zero for $n=2m-1$ and for $n=2m$ equal to $(n-1)!!$. Thus $$ g(n) = \frac{(-1)^{n}+1}{2} (n-1)!! $$ If, however, there is no typo in your expression and that is you meant $$ \tilde{g}(n) =(-1)^{n-1} \sum_{r=0}^n \binom{n}{r} f(r) = (-1)^{n-1} \mathbb{E}((Z+1)^n) $$ But $Z+1$ is again a normal variable with $\mu=2$ and $\sigma^2=1$, hence $$ g(n) = (-1)^{n-1} n! [t]^n \mathrm{e}^{2t + t^2/2} = (-1)^{n-1} \sum_{j \geqslant 0 } \frac{n}{2j} 2^{n-2j} \frac{(2j)!}{2^j j!} $$