Semantic clustering based deduction learning for image recognition and classification

The paper proposes a semantic clustering based deduction learning by mimicking the and thinking process of human brains. Human beings can make judgments on experience cognition, as result, no one would recognize an unknown animal car. Inspired this observation, we propose to train deep models using prior that guide learn with ability deducing summarizing from classification attributes, such cat belonging animals while car pertaining vehicles. proposed approach realizes high-level in space, enabling model deduce relations among various classes during process. In addition, introduces random search for opposite labels ensure smooth distribution robustness classifiers. is supported theoretically empirically through extensive experiments. We compare performance across state-of-the-art classifiers popular benchmarks, generalization verified adding noisy labeling datasets. Experimental results demonstrate superiority approach.