N07-2004 |
phonological features to independent
|
feature detectors
|
in a Conditional Random Fields
|
P14-1062 |
as - pects . The range of the
|
feature detectors
|
is limited to the span m of the
|
P14-1062 |
separate n-grams . We visualise the
|
feature detectors
|
in the first layer of the network
|
P14-1062 |
filter m correspond to a linguistic
|
feature detector
|
that learns to recog - nise a
|
D15-1279 |
vector . Let nc be the number of
|
feature detectors
|
. The output of the tree-based
|
P14-1062 |
experiments and we inspect the learnt
|
feature detectors
|
. 5.1 Training In each of the
|
D09-1035 |
nonlinear feature detectors and linear
|
feature detectors
|
in the final layer . As shown
|
D15-1279 |
the output layer and underlying
|
feature detectors
|
, enabling effective structural
|
N07-2004 |
the independently trained MLP
|
feature detectors
|
used in previous work . 2 Conditional
|
P14-1062 |
representation . With a folding layer , a
|
feature detector
|
of the i-th order depends now
|
N07-2004 |
Here we evaluate phonological
|
feature detectors
|
created from MLP phone posterior
|
P14-1062 |
reason higher-order and long-range
|
feature detectors
|
can not be easily incorporated
|
D09-1035 |
autoencoder with stochastic nonlinear
|
feature detectors
|
and linear feature detectors
|
D15-1279 |
design a set of fixed-depth subtree
|
feature detectors
|
, called the tree-based convolution
|
D15-1206 |
networks . By randomly omitting
|
feature detectors
|
from the network during train
|
P14-1062 |
in order to learn fine-grained
|
feature detectors
|
, it is beneficial for a model
|
P14-1062 |
formulation of the network so far ,
|
feature detectors
|
applied to an individual row
|
P14-1062 |
the DCNN is associated with a
|
feature detector
|
or neuron that learns during
|
P14-1062 |
width 7 , for each of the 288
|
feature detectors
|
we rank all 7-grams occurring
|
D13-1176 |
and can be thought of as learnt
|
feature detectors
|
. From the sentence matrix Ee
|