Search for: All records

Abstract Spatial transcriptomics analysis is a powerful approach for dissecting the structure of tissue microenvironment and uncovering the mechanism of cell–cell communications. However, existing technologies are limited by either spatial resolution or gene coverage. Most single-cell-resolution technologies target only a few hundred preselected genes, whose choice plays an important role in the overall analysis. It remains a challenge to optimally design a gene panel to maximize the utility of spatial transcriptomics profiling. To fill this gap, we introduce a novel method, named ReconST, to automatically design optimal gene panels for spatial transcriptomics profiling. ReconST leverages information from existing scRNA-seq data and identifies the optimal subset of genes by using a gated autoencoder. By using a high-coverage mouse brain MERFISH dataset and a fetal lung dataset as the reference benchmarks, we showed that ReconST outperforms existing methods in terms of reconstruction accuracy, spatial pattern preservation, and computing efficiency. As such, ReconST provides a useful and generally applicable tool for optimal gene panel design, which in turn can significantly enhance the utility of spatial transcriptomics profiling in a wide range of biomedical investigations. 

Automated scoring of written constructed responses typically relies on separate models per task, straining computational resources, storage, and maintenance in real-world education settings. We propose UniMoE-Guided, a knowledge-distilled multi-task Mixture-of-Experts (MoE) approach that transfers expertise from multiple task-specific large models (teachers) into a single compact, deployable model (student). The student combines (i) a shared encoder for cross-task representations, (ii) a gated MoE block that balances shared and task-specific processing, and (iii) lightweight task heads. Trained with both ground-truth labels and teacher guidance, the student matches strong task-specific models while being far more efficient to train, store, and deploy. Beyond efficiency, the MoE layer improves transfer and generalization: experts develop reusable skills that boost cross-task performance and enable rapid adaptation to new tasks with minimal additions and tuning. On nine NGSS-aligned science-reasoning tasks (seven for training/evaluation and two held out for adaptation), UniMoE-Guided attains performance comparable to per-task models while using 6x less storage than maintaining separate students, and 87x less than the 20B-parameter teacher. The method offers a practical path toward scalable, reliable, and resource-efficient automated scoring for classroom and large-scale assessment systems. 

Medical images exhibit latent anatomical groupings, such as organs, tissues, and pathological regions, that standard Vision Transformers (ViTs) fail to exploit.While recent work like SBM-Transformer attempts to incorporate such structures through stochastic binary masking, they suffer from non-differentiability, training instability, and the inability to model complex community structure. We present DCMM-Transformer, a novel ViT architecture for medical image analysis that incorporates a Degree-Corrected Mixed-Membership (DCMM) model as an additive bias in self-attention. Unlike prior approaches that rely on multiplicative masking and binary sampling, our method introduces community structure and degree heterogeneity in a fully differentiable and interpretable manner. Comprehensive experiments across diverse medical imaging datasets, including brain, chest, breast, and ocular modalities, demonstrate the superior performance and generalizability of the proposed approach. Furthermore, the learned group structure and structured attention modulation substantially enhance interpretability by yielding attention maps that are anatomically meaningful and semantically coherent. 

As machine learning and artificial intelligence are being integrated into cyber-physical systems, it is becoming important for engineers to know and understand these topics. In particular, sensor data is on the rise in these systems and therefore engineers need to understand which models are appropriate to time-series sensor data and how signal processing can be used with them. The Center for Cyber-Physical Systems (CCPS) at the University of Georgia (UGA) is addressing these issues. Student researchers in the CCPS require skills in these areas. This paper demonstrates a machine learning framework for time-series sensor data that can be used to quickly build, train, and test multiple models on CCPS testbed data. The framework is also a tool that can be used as a tutorial to help student researchers understand the concepts required to be successful in the CCPS. 

Abstract How can the NGO–NGO interactions of Global South organizations be better understood and improved? Global South nongovernmental organizations (NGOs) have been frequently missing from the overall advocacy network. When they are included, global South organizations are less likely to take on brokerage roles that may help them build connections and resources across communities. Instead, if they are included, Global South NGOs often are relegated to less powerful, intra-community brokerage roles. Drawing on Deloffre and Quack’s framework (Chapter 1, this volume), Chapter 6 examines the top Global South NGOs that have been able to overcome exclusionary structures and forge inter-community brokerage connections to other NGOs. A deeper look at these organizations and the structures where they are embedded can help to gain insights into the transformative nature of NGO–NGO interactions. The chapter finds that certain country, community, and organizational factors help some Global South NGOs develop connections outside of their immediate community. A focus on these factors may help innovation and protect against a civil society backlash.