In-context learning (ICL) has proven to be adept at adapting large language models (LLMs) to downstream tasks without parameter updates, based on a few demonstration examples. Prior work has found that the ICL performance is susceptible to the selection of examples in prompt and made efforts to stabilize it. However, existing example selection studies ignore the ethical risks behind the examples selected, such as gender and race bias. In this work, we conduct extensive experiments and discover that (1) example selection with high accuracy does not mean low bias; (2) example selection for ICL may amplify the biases of LLMs; (3) example selection contributes to spurious correlations of LLMs. Based on the above observations, we propose the Remind with Bias-aware Embedding (ReBE), which removes the spurious correlations through contrastive learning and obtains bias-aware embedding for LLMs based on prompt tuning. Finally, we demonstrate that ReBE effectively mitigates biases of LLMs without significantly compromising accuracy and is highly compatible with existing example selection methods.

Single-cell RNA sequencing (scRNA-seq) enables comprehensive investigation of cellular heterogeneity at the transcriptomic level. Nevertheless, the high dimensionality, sparsity, and technical noise inherent to scRNA-seq data pose significant challenges for precise cell type identification. Traditional clustering methods struggle to capture the nonlinear structures of such data, while existing deep learning approaches commonly suffer from inadequate disentanglement of semantic features from technical noise. To mitigate these limitations, we present scDiffAE, a new clustering approach built upon a diffusion autoencoder. It leverages a semantic encoder to extract low-dimensional semantic embeddings, and incorporates a conditional diffusion model that conditions on both semantic and random encoding features to progressively reconstruct the original data, effectively decoupling semantic features from technical noise. To further promote disentanglement and preserve semantic integrity, an auxiliary decoder is introduced to directly reconstruct the input from the semantic embeddings. Evaluations on 11 real-world scRNAseq datasets demonstrate that scDiffAE consistently surpasses seven leading methods across multiple metrics, including ARI, AMI, and NMI. Further differential gene expression analysis validates the reliability of the clustering results, providing a solid foundation for downstream research.

Single-cell multi-modal analysis has made significant progress in recent years. However, the translation and generation of data across modalities remain challenging due to differences in data types, measurement scales, and inherent noise. In this study, we propose a novel method for single-cell cross-modality data generation based on a feature-decoupling conditional diffusion framework. Specifically, the method decouples modality-unique and common features using modality-specific and shared encoders, and then leverages two conditional diffusion models to enable accurate bidirectional translation between modalities without requiring prior cell-type annotations. Experimental results on four real datasets demonstrate that our approach significantly outperforms existing state-of-the-art methods in cross-modality translation tasks, achieving higher Pearson correlation coefficients and AUROC scores for RNA and ATAC generation. The proposed framework offers a promising solution for cross-modality analysis in single-cell research.

Single-cell multi-omics technologies enable the exploration of relationships between different omics layers and the cell-type-specific molecular regulatory mechanisms. However, these technologies face challenges such as data acquisition difficulty, high costs, and limited throughput. Existing cross-modal translation methods also face limitations in handling unpaired data, preserving modality heterogeneity, and modeling complex mapping relationships. To address these issues, we propose scOTD-a bidirectional cross-modal translation framework for unpaired single-cell data, based on optimal transport (OT) and conditional diffusion models. This method overcomes the dependency on paired data or prior knowledge by modeling independent latent spaces, thus preserving the unique features of each modality. It further establishes probabilistic coupling mechanisms using OT to implicitly model cross-modal associations and employs dynamic weight injection into a conditional diffusion model for mutual translation across modalities. Experimental results on four scRNA-seq and scATAC-seq datasets show that scOTD outperforms existing state-of-the-art methods across multiple evaluation metrics, providing a robust and efficient solution for single-cell cross-modal translation.

Text-to-Image (T2I) diffusion models exhibit concerning tendencies to generate harmful imagery that perpetuates social biases and stereotypes, posing significant ethical risks in real-world applications. While existing mitigation approaches predominantly employ black-box methodologies through dataset augmentation or constrained fine-tuning, they face critical limitations, including high data acquisition costs and potential exacerbation of stereotypes during model retraining. Inspired by neuroscience principles where neurological dysfunction often stems from aberrant neural activation patterns, we propose a novel framework, StereoClinic, targeting the root cause of stereotype generation through direct neural intervention. Our solution introduces two synergistic components: Diffusion Deep Taylor Decomposition (DDTD) for precisely localizing stereotype-related neurons via Layer-wise Relevance Propagation (LRP) attribution analysis, and Stereotype Neuron Suppression (SNS) implementing targeted activation damping to neutralize bias propagation. Through extensive empirical evaluations across multiple bias dimensions, we demonstrate that our method achieves significant stereotype mitigation without compromising image quality or requiring additional training data. This neuro-inspired approach establishes a new paradigm for model interpretability and ethical alignment in generative AI systems.

Single-cell multi-omics technologies enable comprehensive molecular profiling, offering insights into cellular heterogeneity and biological mechanisms. However, current cross-modality translation methods struggle with high-dimensional, noisy, and sparse single-cell data. We propose single-cell Diffusion models for Cross-modality Translation (scDCT), a probabilistic framework for bidirectional cross-modality translation in single-cell data, including single-cell RNA sequencing, single-cell assay for transposase-accessible chromatin sequencing, and protein expression. scDCT integrates modality-specific autoencoders with conditional denoising diffusion probabilistic models to map inputs to latent spaces and perform probabilistic translation across modalities. This design captures cell-type heterogeneity, accounts for data sparsity, and models uncertainty during translation. Extensive experiments on eight benchmark datasets demonstrate that scDCT outperforms state-of-the-art methods across paired, unpaired, cross-type, and cross-tissue settings, offering a robust and interpretable solution for single-cell multi-omics integration.

Accurately inferring clonal mutational profiles is essential for understanding intra-tumor heterogeneity and clonal selection during tumor evolution. Single-cell multi-modal genomic data, such as copy numbers and point mutations, can be integrated to deliver multiple views of the clonal mutational patterns. Despite of the fact that integration of single-cell multi-modal data has been extensively explored in existing studies, computational methods specifically developed to integrate copy number and point mutation data of single cells are still highly needed. We introduce a deep joint representation learning framework called scCMP, to accurately identify clonal mutational profiles. scCMP employs hybrid Transformer-CNN architectures and graph convolutional networks to integrate single-cell copy number and point mutation data. By fusing individual and commonality information among the two modalities, it generates meaningful cell embeddings for identifying clonal clusters. We comprehensively evaluate the effectiveness of scCMP on five real single-cell DNA sequencing datasets, and further showcase its good scalability on datasets generated from other omics technologies. The results show scCMP accurately aggregates the cells with similar mutational profiles into a same cluster, and surpasses the state-of-the-art methods, indicating its advantage in integrating single-cell genomic data.

Dropout events and technical noise are pervasive challenges in single-cell RNA sequencing (scRNA-seq) data, often obscuring true gene expression profiles and undermining the reliability of downstream analyses. Existing imputation and denoising methods offer partial relief but frequently struggle with over-smoothing and fail to fully capture the complex heterogeneity of cellular states. To address these limitations, we introduce scSTD, a novel imputation and denoising framework that uniquely combines the Swin Transformer (SwinT) architecture with a latent diffusion model. In scSTD, a deep autoencoder first encodes each cell into a compact latent embedding, which is then modeled via a SwinT-based latent diffusion process designed to learn the rich, multimodal distribution of scRNA-seq data. This integration enables scSTD to accurately recover gene expression profiles while preserving subtle biological variation. By synthesizing realistic latent neighbors for each cell and aggregating their decoded outputs, scSTD achieves high-fidelity imputation and denoising. Comprehensive evaluations on both synthetic and real scRNA-seq datasets demonstrate that scSTD significantly outperforms existing methods in recovering true gene expression profiles and maintaining the topological integrity of cellular landscapes.

Text-to-Image (T2I) has witnessed significant advancements, demonstrating superior performance for various generative tasks. However, the presence of stereotypes in T2I introduces harmful biases that require urgent attention as the T2I technology becomes more prominent. Previous work for stereotype mitigation mainly concentrated on mitigating stereotypes engendered with individual objects within images, which failed to address stereotypes engendered by the association of multiple objects, referred to as *Association-Engendered Stereotypes*. For example, mentioning ''black people'' and ''houses'' separately in prompts may not exhibit stereotypes. Nevertheless, when these two objects are associated in prompts, the association of ''black people'' with ''poorer houses'' becomes more pronounced. To tackle this issue, we propose a novel framework, MAS, to Mitigate Association-engendered Stereotypes. This framework models the stereotype problem as a probability distribution alignment problem, aiming to align the stereotype probability distribution of the generated image with the stereotype-free distribution. The MAS framework primarily consists of the *Prompt-Image-Stereotype CLIP* (*PIS CLIP*) and *Sensitive Transformer*. The *PIS CLIP* learns the association between prompts, images, and stereotypes, which can establish the mapping of prompts to stereotypes. The *Sensitive Transformer* produces the sensitive constraints, which guide the stereotyped image distribution to align with the stereotype-free probability distribution. Moreover, recognizing that existing metrics are insufficient for accurately evaluating association-engendered stereotypes, we propose a novel metric, *Stereotype-Distribution-Total-Variation*(*SDTV*), to evaluate stereotypes in T2I. Comprehensive experiments demonstrate that our framework effectively mitigates association-engendered stereotypes.