Research interests
I am interested in building more reliable and trustworthy deep learning predictors. My vision is that this will ultimately be achieved by having a fine-grained and contextual understanding of generalisation, which will probably be achieved through a combination of empirical results and theoretical work. On the medium term, steps toward this goal should be achieved by leveraging both interpretability, to highlight predictive mechanisms used by models, and uncertainty estimation, which can provide great hints of what signal is being captured by understanding what happens when a model is presented with data it is uncertain about.
Publications
- Fuzzy Logic for Biological Networks as ML Regression: Scaling to Single-Cell Datasets With Autograd
(2022)
Le Constance, Alice Driessen, Nicolas Deutschmann, Maria Rodriguez Martinez
We present the BioFuzzNet module, a fuzzy logic tool to model signal transduction in biological networks.
- Is attention interpretation? A quantitative assessment on sets
(2022)
Jonathan Haab, Nicolas Deutschmann, Maria Rodriguez Martinez
The debate around the interpretability of attention mechanisms is centered on whether attention scores can be used as a proxy for the relative amounts of signal carried by sub-components of data.
- Modelling CAR T cell signalling using prior knowledge networks
(2022)
Alice Driessen, Rocío Castellanos Rueda, Constance Le Gac, Nicolas Deutschmann, Maria Rodriguez Martinez, Sai Reddy
Chimeric antigen receptor (CAR) T cells are a promising new approach in cancer immunotherapy. Their safety, efficacy and phenotype depend heavily on the design of the CAR, which intracellular tail contains up to three domains derived from a range of cellular signalling receptors.
- Attention-based Interpretable Regression of Gene Expression in Histology
(2022)
Mara Graziani, Niccolò Marini, Nicolas Deutschmann, Nikita Janakarajan, Henning Müller, María Rodríguez Martínez
Interpretability of deep learning is widely used to evaluate the reliability of medical imaging models and reduce the risks of inaccurate patient recommendations.
- Accelerating HEP simulations with Neural Importance Sampling
(2021)
Nicolas Deutschmann, Niklas Götz
Virtually all high-energy-physics (HEP) simulations for the LHC rely on Monte Carlo using importance sampling by means of the VEGAS algorithm.
[...]