Since individual training examples affect multiple memory locations,
we use a simple technique for retrieving from memory when
deciding whether to shoot or to pass. We round
to the nearest
for which Mem[
] is defined, and then take
as the value of
. Thus, each
Mem[
] represents
for
. Notice that retrieval is much simpler when
using this technique than when using kNN or kernel regression: we look
directly to the closest fixed memory position, thus eliminating the
indexing and weighting problems involved in finding the k closest
training examples and (possibly) scaling their results. We used this
retrieval technique throughout our experiments, concentrating the
trickiness of our function learning at storage time.