comment on recent paper by Maas and Orponen

[Pekka Orponen]

Dear connectionists:

In a recent message to this list, C. Omlin mentions a result
of ours about implementing finite state automata in sigmoidal
neural nets, and lists a number of related papers.

We are familiar with the work listed, at least the journal papers,
and it is cited in our paper (at least in the full version;
some references may have been omitted for lack of space from
the NIPS version).

The difference is that we wanted a _noise-tolerant_ simulation
of automata in sigmoidal _first-order_ networks. The result is
really simple, and is included in the paper only because we
could not find a source for it in the literature.

As far as I recall, the simulations of Frasconi et al. don't
pay attention to the noise tolerance requirement, and those of
Omlin et al. use higher-order networks.

Regards,
   Pekka Orponen