Long-tailed Visual Recognition via Gaussian Clouded Logit Adjustment

<p>Long-tailed data is still a big challenge for deep neural networks, even though they have achieved great success on balanced data. We observe that vanilla training on long-tailed data with cross-entropy loss makes the instance-rich head classes severely squeeze the spatial distribution of the tail classes, which leads to difficulty in classifying tail class samples. Furthermore, the original cross-entropy loss can only propagate gradient short-lively because the gradient in softmax form rapidly approaches zero as the logit difference increases. This phenomenon is called softmax saturation. It is unfavorable for training on balanced data, but can be utilized to adjust the validity of the samples in long-tailed data, thereby solving the distorted embedding space of long-tailed problems. To this end, this paper therefore proposes the Gaussian clouded logit adjustment by Gaussian perturbing different class logits with varied amplitude. We define the amplitude of perturbation as cloud size and set relatively large cloud sizes to tail classes. The large cloud size can reduce the softmax saturation and thereby making tail class samples more active as well as enlarging the embedding space. To alleviate the bias in the classifier, we accordingly propose the class-based effective number sampling strategy with classifier re-training. Extensive experiments on benchmark datasets validate the superior performance of the proposed method.</p><br>

Long-tailed Visual Recognition via Gaussian Clouded Logit Adjustment

<p>Long-tailed data is still a big challenge for deep neural networks, even though they have achieved great success on balanced data. We observe that vanilla training on long-tailed data with cross-entropy loss makes the instance-rich head classes severely squeeze the spatial distribution of the tail classes, which leads to difficulty in classifying tail class samples. Furthermore, the original cross-entropy loss can only propagate gradient short-lively because the gradient in softmax form rapidly approaches zero as the logit difference increases. This phenomenon is called softmax saturation. It is unfavorable for training on balanced data, but can be utilized to adjust the validity of the samples in long-tailed data, thereby solving the distorted embedding space of long-tailed problems. To this end, this paper therefore proposes the Gaussian clouded logit adjustment by Gaussian perturbing different class logits with varied amplitude. We define the amplitude of perturbation as cloud size and set relatively large cloud sizes to tail classes. The large cloud size can reduce the softmax saturation and thereby making tail class samples more active as well as enlarging the embedding space. To alleviate the bias in the classifier, we accordingly propose the class-based effective number sampling strategy with classifier re-training. Extensive experiments on benchmark datasets validate the superior performance of the proposed method.</p><br>

Vertical interface augmented tunability of scattering spectra in ferromagnetic microwire/silicone rubber metacomposites

Previously, we have demonstrated a viable approach based on microstructural and topological modulation of periodically arranged elements to program wave scattering in ferromagnetic glass-coated microwire metacomposites. In order to fully exploit the intrinsic structure of the composite, here, we implement the concept of composites plainification by an in-built vertical interface on randomly dispersed short-cut microwires allowing the adjustment of electromagnetic properties to a larger extent. Such interface was modified through arranging wires with different internal structures in two separated regions and by alternating these regions through wire concentration variations associated with polarization differences across the interface. When the wire concentration was equal in both regions, two well-defined transmission windows with varied amplitude and bandwidth were generated. Wire concentration fluctuations resulted in strong scattering changes ranging from broad passbands to pronounced stopbands, demonstrating the intimate relationship between wire content and space charge variations at the interface. This provides a new method to rationally exploit interfacial effects and microstructural features of microwire metacomposites. Moreover, the advantages of enabling tunable scattering spectra by merely 0.053 vol.% of fillers and simple structure make the proposed plainification strategy instrumental to designing filters with broadband frequency selectivity.