Missing critical modalities in medical imaging poses significant challenges for AI-driven diagnostic systems, particularly in scenarios where limited modalities must suffice for downstream tasks. Existing approaches often fail to fully leverage privileged features available only at training or address the information gap between privileged and limited modalities, resulting in suboptimal performance. To address this, we propose a unified, dual-stage Disentanglement-AligNmenT framEwork (DANTE), which uses InformationTheoretic Regularization and Cross-Modal Reconstruction to decompose full-modality information into alignable and privileged-exclusive components. In the first stage, a self-supervised pre-training strategy based on cross-modal reconstruction acts as a proxy task to implicitly incentivize disentangled representations. In the second stage, we present an information-theoretic regularization to explicitly maximize the transfer of privileged knowledge through two novel modules: (1) a Mutual Alignment Module that employs multilevel bidirectional alignment between limited-modality features and alignable features, enhancing cross-modal representation consistency; (2) a Privileged Compaction Module that restricts the privileged-exclusive information flow, promoting the integration of task-relevant content into alignable representations. Experimental results on three challenging medical datasets demonstrate that DANTE achieves state-of-the-art performance, demonstrating its effectiveness in leveraging privileged guidance under modality scarcity, and exhibits broad applicability across diverse medical imaging scenarios.

In recent years, remote laboratories have become integral to modern education, offering flexibility and accessibility compared to traditional, in-person labs. Integrating AI-powered assistance into remote labs has the potential to give them an edge by providing personalized learning experiences. This paper explores an innovative approach to promoting independent learning and critical thinking by embedding AI-driven support, using OpenAI's GPT-4 model, into a remote Field Programmable Gate Array (FPGA) laboratory. Through a web-based code editor, students write SystemVerilog programs and receive tailored assistance from the AI, while their designs are deployed on a Terasic DEl-SoC FPGA development board with real-time feedback via a live camera feed. The study, which involved students from an advanced digital design course interacting with the AI assistant, revealed strong engagement and positive feedback. Preliminary results indicate that AI-powered guidance can meaningfully boost student involvement, providing a scalable and effective framework for fostering active learning in engineering education.

Social Networking Services (SNS) are one of the most widely used web applications and have a profound influence on the real world. Network structure analysis plays a crucial role in the optimization of the functionalities of SNS. In particular, subgraph statistics is widely used as a fundamental method to reveal the basic features and patterns of networks. However, one of the main challenges in data collection is how to perform effective analysis without violating users’ privacy. Existing privacy-preserving subgraph statistics methods mainly rely on enumeration, which are often inefficient or require multiple rounds of interactions. It is difficult to meet the needs of rapidly evolving social network services. To address this problem, we propose a set of new algorithms that aim to improve the efficiency and accuracy of subgraph counting under privacy preservation. By focusing on two typical network structures, the Random network and Barabási-Albert network, our approach is able to efficiently analyse the network structure while ensuring individual privacy. These algorithms perform with high accuracy in experiments, especially in evaluating triangles in BA networks, when compared with the state-of-the-art algorithm.