Time Series Representation Learning 4 Health @ICLR on Twitter

“Interested in representation learning from time series data + medical applications? We are happy to announce the 1st hybrid workshop on time series representation learning for health #TSRL4H at ICLR'23 (@iclr_conf)! Website: https://t.co/pSMQBGZx52”

6 Principal Investigators (m/f/d) as Hector Endowed ELLIS Fellows in Tübingen

The ELLIS mission is to create a diverse European network that promotes research excellence and advances breakthroughs in AI, as well as a pan-European PhD program to educate the next generation of AI researchers. ELLIS also aims to boost economic growth in Europe by leveraging AI technologies.

Invariance Learning in Deep Neural Networks with Differentiable Laplace Approximations

Data augmentation is commonly applied to improve performance of deep learning by enforcing the knowledge that certain transformations on the input preserve the output. Currently, the data augmentation parameters are chosen by human effort and costly cross-validation, which makes it cumbersome to apply to new datasets. We develop a convenient gradient-based method for selecting the data augmentation without validation data during training of a deep neural network. Our approach relies on phrasing data augmentation as an invariance in the prior distribution on the functions of a neural network, which allows us to learn it using Bayesian model selection. This has been shown to work in Gaussian processes, but not yet for deep neural networks. We propose a differentiable Kronecker-factored Laplace approximation to the marginal likelihood as our objective, which can be optimised without human supervision or validation data. We show that our method can successfully recover invariances present in the data, and that this improves generalisation and data efficiency on image datasets.

Assistant Professors (Tenure Track) of Computer Science – Visualization

Anshul Kundaje ([email protected]) on Twitter

“@gxr I seriously think something needs to be done about these open access fees. They are crazy high and I don't think I have seen a numerical justification for these numbers, especially when nobody really cares about the paper printed version of a journal article any more.”

Steven Salzberg 💙💛 on Twitter

“I was curious about how one could possibly use machine learning to improve metagenomic assembly, but Olga Mineeva is describing a very clever idea about how to do this at #biodata22 right now @gxr”

“Michael” on Twitter

“Back to sc perturbations with @stefangstark on using neural optimal transport to study cell responses, given the challenge that we never can see a cell with and without perturbation. Work with @gxr at ETH Zurich #biodata22”