David Klindt, PhD -  Postdoctoral Research Associate, Stanford University, Machine Learning and Computational Neuroscience 

“Natural Visual Intelligence – Learning, Inference and Generalisation of Visual Representations in Brains and Machines”

How do animals make sense of the visual world? This presentation provides a comprehensive overview of my research agenda addressing this fundamental question. At a high level, I am interested in robust and intelligent perception. I want to find out how animals can learn structured representations in a dynamically changing world, how they are able to robustly infer and compose these representations to generalise to new situations, and how we can build Artificial Intelligence (AI) models with the same efficient learning and inference mechanisms. Building on a foundation of both computational neuroscience and machine learning, examples of my past research illustrate the transfer of pivotal concepts, such as equivariant neural representations, from visual neurons to the navigation system. This framework enhances model performance and provides insights into the underlying biological mechanisms, for instance, by identifying functional cell types. Moreover, in an example from my past work in machine learning, I demonstrate how developmental psychology inspires a solution to the hard problem of learning object-centric visual representations of the world. These endeavours stand as testament to the interplay between biological insight and technological advancement. In my current research, I employ psychophysics to quantitatively measure the interpretability of neural representations, providing a rigorous basis to test the recent superposition hypothesis in both convolutional neural networks and the visual cortex. Looking forward, I outline a project in hierarchical visual inference, illustrating how it aligns with my overarching research vision to gain deeper insights into psychological and neural processes, ultimately leading to the development of more advanced machine learning models for visual perception. Thus, I propose the Natural Visual Intelligence Lab with a research agenda that offers new insights into neural representations and catalyses the development of the next generation of more robust and interpretable AI models, reshaping the future of intelligent systems.

David Klindt received his Ph.D. in Machine Learning & Computational Neuroscience from the University of Tübingen and is currently a Postdoctoral Research Associate at Stanford University. His research focuses on understanding how the brain processes dynamic environmental stimuli using modern machine learning techniques. He aims to leverage these insights to create advanced machine learning models capable of flexibly and robustly extracting meaningful information from high-dimensional sensory data in diverse settings. David is particularly interested in computational neuroscience, machine learning and computer vision, with recent work encompassing generative models, disentanglement, domain adaptation, robustness, and topological data analysis.