Differentiable self-supervised clustering with intrinsic interpretability.

Self-supervised clustering has garnered widespread attention due to its ability to discover latent clustering structures without the need for external labels. However, most existing approaches on self-supervised clustering lack of inherent interpretability in the data clustering process. In this paper, we propose a differentiable self-supervised clustering method with intrinsic interpretability (DSC2I), which provides an interpretable data clustering mechanism by reformulating clustering process based on differentiable programming. To be specific, we first design a differentiable mutual information measurement to explicitly train a neural network with analytical gradients, which avoids variational inference and learns a discriminative and compact representation. Then, an interpretable clustering mechanism based on differentiable programming is devised to transform fundamental clustering process (i.e., minimum intra-cluster distance, maximum inter-cluster distance) into neural networks and convert cluster centers to learnable neural parameters, which allows us to obtain a transparent and interpretable clustering layer. Finally, a unified optimization method is designed, in which the differentiable representation learning and interpretable clustering can be optimized simultaneously in a self-supervised manner. Extensive experiments demonstrate the effectiveness of the proposed DSC2I method compared with 16 clustering approaches.

Gut microbial diversity moderates polygenic risk of schizophrenia.

Background: Schizophrenia (SCZ) is a heritable disorder with a polygenic architecture, and the gut microbiota seems to be involved in its development and outcome. In this study, we investigate the interplay between genetic risk and gut microbial markers. Methods: We included 159 first-episode, drug-naïve SCZ patients and 86 healthy controls. The microbial composition of feces was characterized using the 16S rRNA sequencing platform, and five microbial α-diversity indices were estimated [Shannon, Simpson, Chao1, the Abundance-based Eoverage Estimator (ACE), and a phylogenetic diversity-based estimate (PD)]. Polygenic risk scores (PRS) for SCZ were constructed using data from large-scale genome-wide association studies. Effects of microbial α-diversity, microbial abundance, and PRS on SCZ were evaluated via generalized linear models. Results: We confirmed that PRS was associated with SCZ (OR = 2.08, p = 1.22×10-5) and that scores on the Shannon (OR = 0.29, p = 1.15×10-8) and Simpson (OR = 0.29, p = 1.25×10-8) indices were inversely associated with SCZ risk. We found significant interactions (p < 0.05) between PRS and α-diversity indices (Shannon, Simpson, and PD), with the effects of PRS being larger in those exhibiting higher diversity compared to those with lower diversity. Moreover, the PRS effects were larger in individuals with a high abundance of the genera Romboutsia, Streptococcus, and Anaerostipes than in those with low abundance (p < 0.05). All three of these genera showed protective effects against SCZ. Conclusion: The current findings suggest an interplay between the gut microbiota and polygenic risk of SCZ that warrants replication in independent samples. Experimental studies are needed to determine the underpinning mechanisms.

Multitask Image Clustering via Deep Information Bottleneck.

Multitask image clustering approaches intend to improve the model accuracy on each task by exploring the relationships of multiple related image clustering tasks. However, most existing multitask clustering (MTC) approaches isolate the representation abstraction from the downstream clustering procedure, which makes the MTC models unable to perform unified optimization. In addition, the existing MTC relies on exploring the relevant information of multiple related tasks to discover their latent correlations while ignoring the irrelevant information between partially related tasks, which may also degrade the clustering performance. To tackle these issues, a multitask image clustering method named deep multitask information bottleneck (DMTIB) is devised, which aims at conducting multiple related image clustering by maximizing the relevant information of multiple tasks while minimizing the irrelevant information among them. Specifically, DMTIB consists of a main-net and multiple subnets to characterize the relationships across tasks and the correlations hidden in a single clustering task. Then, an information maximin discriminator is devised to maximize the mutual information (MI) measurement of positive samples and minimize the MI of negative ones, in which the positive and negative sample pairs are constructed by a high-confidence pseudo-graph. Finally, a unified loss function is devised for the optimization of task relatedness discovery and MTC simultaneously. Empirical comparisons on several benchmark datasets, NUS-WIDE, Pascal VOC, Caltech-256, CIFAR-100, and COCO, show that our DMTIB approach outperforms more than 20 single-task clustering and MTC approaches.

Cross-Modal Clustering With Deep Correlated Information Bottleneck Method.

Cross-modal clustering (CMC) intends to improve the clustering accuracy (ACC) by exploiting the correlations across modalities. Although recent research has made impressive advances, it remains a challenge to sufficiently capture the correlations across modalities due to the high-dimensional nonlinear characteristics of individual modalities and the conflicts in heterogeneous modalities. In addition, the meaningless modality-private information in each modality might become dominant in the process of correlation mining, which also interferes with the clustering performance. To tackle these challenges, we devise a novel deep correlated information bottleneck (DCIB) method, which aims at exploring the correlation information between multiple modalities while eliminating the modality-private information in each modality in an end-to-end manner. Specifically, DCIB treats the CMC task as a two-stage data compression procedure, in which the modality-private information in each modality is eliminated under the guidance of the shared representation of multiple modalities. Meanwhile, the correlations between multiple modalities are preserved from the aspects of feature distributions and clustering assignments simultaneously. Finally, the objective of DCIB is formulated as an objective function based on a mutual information measurement, in which a variational optimization approach is proposed to ensure its convergence. Experimental results on four cross-modal datasets validate the superiority of the DCIB. Code is released at https://github.com/Xiaoqiang-Yan/DCIB.