... DIKU¹

DIKU: Datalogisk Institut Københavns Universitet, Department of Computer Science University of Copenhagen

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... compiler²

In gcc it is possible to get signals when a overflow occurs with -ftrapv, but unfortunately some optimized pointer arithmetic in gcc makes integer overflow and breaks this functionality (see http://gcc.gnu.org/bugzilla/show_bug.cgi?id=1823).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... aliasing³

C and C++ automaticly thinks that data reached from two different pointers could be the same. This makes for safe but slow code (FORTRAN assumes the opposite, which makes for fast unsafe code).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... layer⁴

Actually this is a past the end pointer, which points to the layer one position past the last layer. But it is easier to think of it as a pointer to the last layer.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... epochs⁵

During one epoch each of the training pairs are trained for one iteration.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... 0.75⁶

The actual connection rate may be different, because some connections should always be in place.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.