Deep neural networks (DNNs) are demonstrated to be vulnerable to adversarial examples. Adversarial training is mainstrem method to improve adversarial robustness of DNNs, which augments the training set with adversarial examples and adopts adversarial regularization loss to improve the robustness of DNNs. Existing adversarial training methods are facing the challenge to balance the accuracy and robustness.

 

Deep neural networks (DNNs) are demonstrated to be vulnerable to adversarial examples. Adversarial training is mainstrem method to improve adversarial robustness of DNNs, which augments the training set with adversarial examples and adopts adversarial regularization loss to improve the robustness of DNNs. Existing adversarial training methods are facing the challenge to balance the accuracy and robustness.

Understanding the dynamics of cyber threats is crucial for today's digital defenses. Researchers from Nanjing University, Jiangsu University of Science and Technology, and Southeast University have discovered that optimizing the execution time of malware in sandbox environments can significantly enhance the completeness and quality of cyber threat intelligence (CTI) data.

New study highlights new ways to produce urea electrochemically, using electricity and waste gases.

A new Children’s Hospital of Philadelphia study examines the relationship between parenting factors and gaming disorder in young children with ADHD. Findings from the study will be presented during the Pediatric Academic Societies (PAS) 2026 Meeting, taking place April 24-27 in Boston.

AI-based “scribes” are increasingly being adopted in emergency medicine to reduce the administrative burden on clinicians by automatically recording and summarizing patient interactions. While these tools may improve documentation efficiency and help address clinician burnout, UVA data science expert Tom Hartvigsen cautions that they also raise important concerns about care quality, system incentives, and patient data privacy.

Hartvigsen notes that because efficiency is easier to measure than quality of care, AI systems may unintentionally steer healthcare delivery toward speed and cost reduction rather than clinical excellence. He also highlights risks related to patient data being processed by generative AI systems, including the potential for sensitive information to be retained or exposed in unintended contexts.

As hospitals weigh adoption decisions, Hartvigsen emphasizes that patients and providers are still actively navigating how AI scribes should be governed, and encourages transparency about whether and how patient data are used in AI systems.

Researchers at The Hong Kong University of Science and Technology (HKUST) have discovered that in perovskite solar cells, conventional passivation strategies mainly act on the film surface and struggle to reach buried interfaces, much like a superficial “external dressing” that cannot repair deep microstructural defects formed during film growth. In particular, the rapid evaporation of solvents during film formation inevitably leads to voids and nanoscale grain-boundary grooves at the bottom interface of perovskite films. These long-overlooked defects severely hinder charge transport and trigger interfacial failure during device scaling and operation, becoming a critical bottleneck for efficiency, stability, and large-area manufacturing.