[CANCELED] Mapping Social Dynamics through Networks: Unveiling Complexity with Statistical Validation

• Daniele Cirulli

Online social networks shape how we access information, form opinions, and engage in public discourse. However, they also pose risks, such as echo chambers and polarization, particularly exacerbated by their instrumental use and the rise of AI-generated content. The vast amount of available data has transformed social science into a more quantitative discipline, with network analysis serving as a key tool to structure social interactions. However, the complexity of these networks makes it challenging to extract meaningful patterns from big data interactions. Statistical methods based on statistical mechanics can filter noise, allowing for unbiased analysis. By combining validated network analysis with natural language processing, we can better understand social patterns in the evolving social media landscape.

Dicussion

A theory of creativity in Convolutional Diffusion Models

• Luca Maria Del Bono

Diffusion models have become the dominant approach for image generation. However, despite their widespread use, they remain poorly understood, and their ability to generate truly novel images—i.e., their "creativity"—still lacks a solid theoretical foundation. In this talk, I will explore this issue by presenting the paper "An Analytic Theory of Creativity in Convolutional Diffusion Models" by Kamb and Ganguli. I will begin with an overview of how diffusion models function, then discuss key open questions regarding their generative capabilities. Finally, I will introduce a theoretical framework that explains creativity in convolutional diffusion networks in terms of equivariance and locality.

Discussion