Multimodal space representation driven by self-evaluation of predictability | IEEE Conference Publication | IEEE Xplore