Deep neural networks (DNNs), despite their impressive ability to generalize over-capacity networks, often rely heavily on malignant bias as shortcuts instead of task-related information for discriminative tasks. To address this problem, recent studies utilize auxiliary information related to the bias, which is rarely obtainable in practice, or sift through a handful of bias-free samples for debiasing. However, the success of these methods is not always guaranteed due to the unfulfilled presumptions. In this paper, we propose a novel method, Contrastive Debiasing via Generative Bias-transformation (CDvG), which works without explicit bias labels or bias-free samples. Motivated by our observation that not only discriminative models but also image translation models tend to focus on the malignant bias, CDvG employs an image translation model to transform one bias mode into another while preserving the task-relevant information. Additionally, the bias-transformed views are set against each other through contrastive learning to learn bias-invariant representations. Our method demonstrates superior performance compared to prior approaches, especially when bias-free samples are scarce or absent. Furthermore, CDvG can be integrated with the methods that focus on bias-free samples in a plug-and-play manner for additional enhancements, as demonstrated by diverse experimental results.

Ruth is a mathematician and data scientist specialising in operations research, machine learning and statistics. She holds a doctorate in mathematics for her research on dynamic resource allocation. She has nearly 20 years of project management, machine learning, programming, and solution development experience in the health, education, and private sectors. At Fireball, she leads the development team building an early bushfire notification platform that uses deep learning to detect fires within minutes of ignition. We're putting the people of Queensland front and centre to support Queensland AI Hub's mission - connecting Queensland's AI ecosystem.

Nothing is more associated with death and destruction than fire. It can produce a primal fear in all of us. So, when a university professor in Seoul, South Korea, challenged his class to use data to find solutions for complicated real-world problems, one student suggested analyzing information held by the city's Fire Department. The idea was to predict the probability of fires so that authorities could take action to make the city safer for its more than 9 million residents. Hongik University's Professor Jae Seung Lee and his students used artificial intelligence (AI) and machine learning (ML) algorithms to develop a new model that can now do just that.

Whenever I get a chance, I pick up my camera and spend hours capturing the immaculate beauty of Big Basin Redwood State Park. It's been my favorite pastime for years. Giant sequoias, which are the world's largest single trees and largest living thing by volume, always help me understand our connection with something bigger than ourselves. Last year, those towering trees were being turned into ashes by wildfires. Watching the destruction of centuries-old trees, I challenged myself to find a solution to stop this colossal loss. When I was 15 years old, I had started a nonprofit organization, Raindrop US, to help raise awareness about the California drought.

It's been a banner year for cyber criminals. International cybersecurity disasters such as the WannaCry and Goldeneye ransomware attacks impacted thousands of people around the globe and illustrated just how tenuous a grasp most organizations hold on their security. Then there's the Equifax debacle, which impacted about 1 in 3 Americans. And if the CIA can't protect its own data from ending up on Wikileaks, what chance do the rest of us stand against ever-more-sophisticated hackers? In theory, artificial intelligence can provide new forms of protection against nefarious actors.