Pretrained large-scale AI models need to ‘forget’ specific information for privacy and computational efficiency, but no methods exist for doing so in black-box vision-language models, where internal details are inaccessible. Now, researchers from Japan addressed this issue through an innovative strategy based on latent context sharing, successfully getting an image classifier to forget multiple classes it was trained on. Their findings could expand the use cases of large-scale AI models while safeguarding end users’ privacy.

Autism spectrum disorder (ASD) is a complex developmental disorder that causes significant social, communication, and behavioral challenges. Early diagnosis, along with suitable therapies, is crucial to alleviating the symptoms. To this end, a team of researchers from the United States of America has developed an advanced screening model to identify toddlers with autism from underserved areas. Their model holds the potential for accessible diagnosis for toddlers from culturally and linguistically diverse backgrounds.

S M Feroj Mahmood’s doctoral dissertation investigates how social ventures can modify their business models to scale internationally while maintaining their commitment to sustainability. According to Mahmood, cultural differences and geopolitical shifts make expansion complex, and to grow internationally, social ventures need to have a deep understanding of local markets, adapt distribution methods, and form strong partnerships. Mahmood will defend his dissertation at the University of Vaasa, Finland, on 11 December 2024. 

In fields ranging from immunology and ecology to economics and thermodynamics, multi-scale complex systems are ubiquitous. They are also notoriously difficult to model. Conventional approaches take either a bottom-up or top-down approach. But in disturbed systems, such as a post-fire forest ecosystem or a society in a pandemic, these unidirectional models can’t capture the interactions between the small-scale behaviors and the system-level properties. SFI External Professor John Harte (UC Berkeley) and his collaborators have worked to resolve this challenge by building a hybrid method that links bottom-up behaviors and top-down causation in a single theory. 

For the second consecutive year, three faculty members in the Department of Biology have received National Institutes of Health Maximizing Investigators’ Research Awards (MIRA) recognizing their sustained research potential in bacterial and applied phage biology, neurobiology, and behavioral and evolutionary biology.