Yes, since $M$ is a natural number ("the number of simplified input features"). In the paper, this is in the context of a definition
$$
g(z') = \phi_0 + \sum_{i=1}^M \phi_iz'_i
$$
where $z'$ is being indexed by $i$ for $1 \le i \le M$. The calculation is adding up $\phi_0$ and some of the other $\phi_i$ values, with $z'$ telling us which ones to include in the total.
More generally, people write $B^A$ for the set of functions from $A$ to $B$. If you like, you can think of $\{0,1\}^M$ as meaning the same as $\{0,1\}^{\{1, 2, \dots, M\}}$, which is the set of functions that take in a number between $1$ and $M$, and spit out either $0$ or $1$. If $f$ is a function like that, then you can make a vector of $0$s and $1$s:
$$
(f(1), f(2), f(3), \dots, f(M)).
$$
And also, if you had a vector of that length, you could use it to define a function, just by indexing into the vector. So the function and the vector are really two different views of the same data. Mathematicians often switch freely between these viewpoints.