Considering 20 face images from Caltech 101, we examine setting of sparseness parameters; unless otherwise stated, b = 102, f = 10−5 and h = 10−5. Parameters h and f are varied in (a) and (b), respectively. In (c), we set e = 10−6 and f = 10−6 and make hidden units unconstrained to test the influence of parameter b on the model's sparseness. In all of cases, we show the Layer-2 filters (ordered as above) and an example reconstruction.