set no_bias in deconv layer in fcnxs example

[zhaw]

In recent update, deconvolution layer got bias parameter but this example hasn't been updated yet. If you run this example now, it returns an error. #1619

[tqchen]

@tornadomeet

[tornadomeet]

@zhaw thanks for pointing this out.
@tqchen ,becasue in deconvolution operator, we generally not use the bias term, so in my original implemention of deconvolution, the default setting is use no bias(like caffe), but @piiswrong has changed the default setting 27a3382 (which i do not agree )
@zhaw so you can refer #1477.
and could you please update this pr not to add bias term in deconv(beacuase it used for upsampling), we only need to change symbol_fcnxs.py in which add no_bias=True in all Deconvolution?
such as https://github.com/dmlc/mxnet/blob/master/example/fcn-xs/symbol_fcnxs.py#L140 https://github.com/dmlc/mxnet/blob/master/example/fcn-xs/symbol_fcnxs.py#L160 https://github.com/dmlc/mxnet/blob/master/example/fcn-xs/symbol_fcnxs.py#L175 https://github.com/dmlc/mxnet/blob/master/example/fcn-xs/symbol_fcnxs.py#L182
and do not change init_fcnxs.py, thanks!

[zhaw]

@tornadomeet thanks for your explanation!

[piiswrong]

@tornadomeet I think we can change deconv back to no_bias =true for default.

BTW, is the gradient correctly computed for bias in deconv?

[tornadomeet]

@piiswrong yes, when using bias in deconv, i think the gradient is computed correctly.
@zhaw so with agreed with @piiswrong , could you help do this: change the default no_bias setting from false to true in deconv, which in https://github.com/dmlc/mxnet/blob/master/src/operator/deconvolution-inl.h#L52 ? Then we will not need to change any code in the fcn-xs example. thank you for your patience!

[zhaw]

@tornadomeet no problem, glad I can help.

[tqchen]

is this PR good to be merged. please squash the commits as in http://mxnet.readthedocs.org/en/latest/contribute.html#how-to-combine-multiple-commits-into-one