I will present work on how kernel-based opinion dynamics models, where a kernel governs how individuals update their beliefs based on neighbors, can be learned from data. Focusing on synchronous, discrete-time stochastic models. We implement an expectation-maximization (EM) algorithm that reliably recovers ground truth kernels from networked time series data of real-valued opinions. Our method performs well across a range of known kernels and is extended to handle non-differentiable kernels (e.g., bounded confidence) using gradient-free optimization. We also present a non-parametric variant using a multilayer perceptron to flexibly learn kernel shapes. Our approach accurately infers diverse kernel types and parameterizations, demonstrating its potential for fitting opinion dynamics models to empirical data. It supports both testing hypothesized kernels and learning them directly from data, regardless of their form or differentiability.
Network science is inherently interdisciplinary, with collaboration across disciplines as its norm. Rather than being confined to a single domain, it offers a versatile toolkit of methods that can be applied across fields, presenting both opportunities and challenges for young researchers. As an early career interdisciplinary researcher, questions arise about whether to find with a specific niche—such as biology, social science, or economics—or to establish oneself as a methodological expert spanning multiple areas. While domain expertise characterizes traditional specialization, pursuing a generalist approach may risk overlooking important details. Historically, researchers and academics have been viewed as specialists, with deep understanding in a narrow area of expertise. In pursuing a more generalist path, one risks lacking the depth needed to grasp the minutiae of any single field. Is it irresponsible to research in a domain in which one is not an expert? Can and should one rely on collaboration to fill in gaps in domain expertise? Finally, is it advisable for a young researcher to strive to find a niche?
Should I specialize or should I stay generalist? In this seminar, we hosted William Thompson (University of Vermont, Burlington, USA), who presented his work on learning kernel-based opinion dynamics models from data and led a discussion on the challenges of navigating an interdisciplinary research path.
William presented a method to learn how individuals update opinions in networks using data. Using EM expectation-maximization and neural nets, his approach recovers different kernel types directly from opinion time series.
In the discussion, we reflected on the risks and responsibilities of working outside one’s core discipline. He emphasized the role of collaboration in bridging knowledge gaps. The key question is: how do we grow our expertise if we stay confined to our niche?