Search

Pseudo-label-based semi-supervised learning (SSL) algorithms trained on a class-imbalanced set face two cascading challenges: 1) Classifiers tend to be biased towards majority classes, and 2) Biased pseudo-labels are used for training. It is difficult to appropriately re-balance the classifiers in SSL because the class distribution of an unlabeled set is often unknown and could be mismatched with that of a labeled set. We propose a novel class-imbalanced SSL algorithm called class-distribution-mismatch-aware debiasing (CDMAD). For each iteration of training, CDMAD first assesses the classifier's biased degree towards each class by calculating the logits on an image without any patterns (e.g., solid color image), which can be considered irrelevant to the training set. CDMAD then refines biased pseudo-labels of the base SSL algorithm by ensuring the classifier's neutrality. CDMAD uses these refined pseudo-labels during the training of the base SSL algorithm to improve the quality of the representations. In the test phase, CDMAD similarly refines biased class predictions on test samples. CDMAD can be seen as an extension of post-hoc logit adjustment to address a challenge of incorporating the unknown class distribution of the unlabeled set for re-balancing the biased classifier under class distribution mismatch. CDMAD ensures Fisher consistency for the balanced error. Extensive experiments verify the effectiveness of CDMAD., Comment: CVPR 2024

Existing semi-supervised learning (SSL) algorithms typically assume class-balanced datasets, although the class distributions of many real-world datasets are imbalanced. In general, classifiers trained on a class-imbalanced dataset are biased toward the majority classes. This issue becomes more problematic for SSL algorithms because they utilize the biased prediction of unlabeled data for training. However, traditional class-imbalanced learning techniques, which are designed for labeled data, cannot be readily combined with SSL algorithms. We propose a scalable class-imbalanced SSL algorithm that can effectively use unlabeled data, while mitigating class imbalance by introducing an auxiliary balanced classifier (ABC) of a single layer, which is attached to a representation layer of an existing SSL algorithm. The ABC is trained with a class-balanced loss of a minibatch, while using high-quality representations learned from all data points in the minibatch using the backbone SSL algorithm to avoid overfitting and information loss.Moreover, we use consistency regularization, a recent SSL technique for utilizing unlabeled data in a modified way, to train the ABC to be balanced among the classes by selecting unlabeled data with the same probability for each class. The proposed algorithm achieves state-of-the-art performance in various class-imbalanced SSL experiments using four benchmark datasets.

Within first-principles density functional theory (DFT) frameworks, accurate but fast prediction of electronic structures of nanoparticles (NPs) remains challenging. Herein, we propose a machine-learning architecture to rapidly but reasonably predict electronic density of states (DOS) patterns of metallic NPs via a combination of principal component analysis (PCA) and the crystal graph convolutional neural network (CGCNN). By applying PCA, one can convert a mathematically high-dimensional DOS image to a low-dimensional vector. The CGCNN plays a key role in reflecting the effects of local atomic structures on the DOS patterns of NPs with only a few of material features (e.g., melting temperature, the number of d electrons, and atomic radius) that are easily obtained from a periodic table. The PCA-CGCNN model is applicable for all pure and bimetallic NPs, in which a handful DOS training sets that are easily obtained with the typical DFT method, such as bulk, slab, and small-sized NPs, are considered. Although there is a small loss of accuracy with the PCA-CGCNN method compared to DFT calculations, the prediction speed is much faster than that of DFT methods and is not nearly as affected by the system sizes of NPs. Our approach not only can be immediately applied to predict electronic structures of actual nanometer scaled NPs to be experimentally synthesized, but also be used to explore correlations between atomic structures and other spectrum image data of the materials (e.g., X-ray diffraction, X-ray photoelectron spectroscopy, and Raman spectroscopy).