Latest Articles for Safety In Technology

A look into how neural networks process information and their implications.

A new approach to safety analysis using piecewise stochastic barrier functions.

A method to train robots for complex tasks while ensuring safety.

A new method improves safety in high-dimensional reachability for autonomous vehicles.

Using Exchange Value to teach AI safe and effective behaviors from humans.

A novel method improving uncertainty estimation in AI models for edge devices.

New method enhances robot driving safety while learning speed.

Introducing an innovative method to enhance robot training and safety.

This article discusses the design of safe neural network controllers for critical systems.

This article discusses enhancing CNNs by leveraging low-frequency information for better resilience against adversarial attacks.

Exploring counterfactual harm constraints for safer decision-making in AI.

Dynamic Model Predictive Shielding enhances safety and efficiency in reinforcement learning.

A method to improve object detection across unseen environments using single-source domain training.

AdvUnlearn enhances image generation safety by combining unlearning with adversarial training.

This article discusses new methods for explaining AI decisions in object detection.

New methods improve detection of outlier samples in mixed data environments.

This article examines the risks of fine-tuning language models for safety.

A new method enables agents to follow safety rules in plain language.

This study explores using smaller models to enhance safety in AI systems.

A novel method for agents to optimize performance while ensuring constraints are met.

This article discusses essential calibration methods for object detectors in critical applications.

Research reveals biases in object detection systems affecting safety in autonomous vehicles.

MaxLin improves CNN verification accuracy and efficiency for safer AI applications.

A new system addresses potential issues in IoT trigger-action rules for smart homes.

New strategies improve deep reinforcement learning against adversarial noise.

Examining the risks and safety measures in fine-tuning language models.

Exploring how cognitive neuroscience can improve AI model comprehension.

A new benchmark suite helps assess reasoning shortcuts in artificial intelligence.

A fresh method for testing language model safety and multilingual skills.

Researchers develop new methods for training robots safely in risky environments.

Research highlights safety neurons' role in enhancing LLM safety and responsibility.

A new approach to enhance model safety through prediction rejection.

Research on magnetic islands enhances plasma stability and disruption prevention in tokamaks.

This article discusses methods for improving AI alignment with various cultures.

A new method helps identify weak points in deep learning models quickly.

Research reveals language models struggle with false reasoning, raising safety concerns.

Research focuses on managing plasma disruptions to improve fusion reactor safety.

CCL ensures neural networks maintain accuracy while learning new tasks.

A tool to analyze and improve computer image recognition errors.

UNRealNet enhances robot navigability in tough terrain using advanced techniques.