My current research concerns formal theory-building (I am a mathematician) in the field of non-digital, 'neuromorphic' computing (NC). Unlike the situation in digital computing, for which a wonderfully transparent and powerful system of interconnected formal theories has been worked out since Alan Turing set the foundation stone in 1936, for NC we painfully lack a general formal grounding. The challenge lies in the condition that NC hardware systems are not digital but evolve in continuous time in spatially organized nonlinear physical substrates (biological neural tissue or engineered new materials). To become practically useful, neuromorphic computing systems must become analyzable, designable, controllable and interpretable across many temporal and spatial scales. Furthermore, when complex physical dynamics in unconventional hardware should become interpretable as 'computing', it is not enough to describe these multiscale dynamics as a physicist would do – in addition to modeling the physical hardware dynamics, one must at the same time also model what these processes 'mean' (which computational tasks they can serve). In the digital computing world, formal theories for capturing the task semantics of computing processes are known, and these formal semantics are key enablers for almost all branches of modern IT technologies. In my talk I explain this problem setting in hopefully understandable ways. While this is all couched in a NC context as far as this talk is concerned, the questions asked are identical to questions that confront other complex systems modelers.

PCA identifies the main factors of variation in a dataset, but it is limited to linear relationships. Normalizing flows can capture more complex patterns, but they lack interpretability.
EOFlows aim to combine the flexibility of normalizing flows with the interpretability of PCA. Using orthogonality regularization, they learn latent dimensions that capture distinct factors of variation, which can be ranked by their explained entropy.
In this talk, Prof. Köthe will explain how EOFlows work and show what they discover in the CelebA portrait dataset. He will also discuss how stable the results are across training runs and how the method compares with existing approaches.

The use of catalysts in an excited state enables a range of chemical transformations that are not accessible by conventional methods. Photoexcitation generates strongly oxidising and reducing agents, which can serve as valuable alternatives to hazardous or costly reagents. Flavins and deazaflavins, prominent natural chromophores (e.g. riboflavin), are characterised by several stable redox and excited states. Consequently, flavin-based photocatalysts are well suited to drive various redox reactions. Recent results from our group and others have shown that flavin species can act as reducing agents, with reducing capabilities comparable to sodium.
In this presentation, the rational design of flavin- and deazaflavin-based systems for photoreductions will be discussed. These systems are based either on a conPET process involving excited flavin radical anions, anionic flavin species and solvated electrons. Applications will be demonstrated, including the generation of phenyl radicals from aryl fluorides, chlorides and bromides, followed by coupling reactions, as well as reductive detriflylation and detosylation. Reductive power of flavin radicals also allows flavin regeneration via solvent (acetonitrile) reduction
in oxidative processes.