# Miscellanea: Frequently Asked Questions

### What's the difference between normal and continuous networks and how
do I make a network one or the other?

Normal networks are the default. These are used for standard
feedforward and simple-recurrent (Elman) networks. In the forward pass
that occurs on each tick in such a network, each group in order will
update its inputs and then immediately its outputs. Therefore, on a
single tick, information can propagate from the first group to the last.
When training a normal network, error derivatives are backpropagated on
each tick and no history of outputs and inputs is needed.

In a *continuous* network, on the other hand, a forward pass is
"synchronous". First all unit outputs are reset to their
*initOutput* if the group's *resetOnExample* flag is
true. Then all of groups calculate their inputs. Finally all of the
groups calculate their outputs based on their inputs. Therefore,
information will not propagate through the entire network on each tick.
It will only propagate the distance of a single link projection.

Continuous networks were given that name because typically a continuous
network would contain groups that continuously integrate their inputs or
outputs over time. However, a continuous network could be given a
*dt* of 1.0 or non-integrating groups, in which case it would not
really be "continuous" per se. Such a network is typically called
recurrent backprop-through-time (RBTT). In Lens it is just considered a
special case of a continuous network.

A network is by default non-continuous. You can make a continuous
network by calling addNet with the
CONTINUOUS network type.

Douglas Rohde
Last modified: Tue Nov 14 00:01:55 EST 2000