What we’re about

This group is a discussion-led forum for practitioners who work with deep learning models in real-world systems and want to better understand how those models actually behave in practice. Rather than focusing on the latest papers or high-level tutorials, the emphasis is on established, widely adopted approaches and unpacking the assumptions, architectural choices, optimisation strategies, and constraints that shape their performance.

Meetups follow a journal-club style format. Each session begins with a short, curated presentation introducing a well-known model or method, using key figures and design decisions as a basis for discussion. The second half of the session is dedicated to open critique and conversation between members, including alternative interpretations, real-world experiences, and domain-specific limitations. The aim is to move beyond “how to use a model” and towards understanding why it works, where it fails, and what it is implicitly assuming about the data.

The group is intended for engineers, data scientists, researchers, and technically curious practitioners applying deep learning across areas such as vision, signal processing, time-series analysis, generative modelling, and physics- or simulation-driven systems. Sessions are run primarily online, with the possibility of occasional in-person events in London if interest develops. The focus is on thoughtful discussion, practical insight, and collective learning rather than product promotion or hype.