ERC FUNDED PROJECTS

The goal of ATOMICAR is to achieve the ultimate sensitivity limit in heterogeneous catalysis: Quantitative measurement of chemical turnover on a single catalytic nanoparticle. Most heterogeneous catalysis occurs on metal nanoparticle in the size range of 3 nm - 10 nm. Model studies have established that there is often a strong coupling between nanoparticle size & shape - and catalytic activity. The strong structure-activity coupling renders it probable that “super-active” nanoparticles exist. However, since there is no way to measure catalytic activity of less than ca 1 million nanoparticles at a time, any super-activity will always be hidden by “ensemble smearing” since one million nanoparticles of exactly identical size and shape cannot be made. The state-of-the-art in catalysis benchmarking is microfabricated flow reactors with mass-spectrometric detection, but the sensitivity of this approach cannot be incrementally improved by six orders of magnitude. This calls for a new measurement paradigm where the activity of a single nanoparticle can be benchmarked – the ultimate limit for catalytic measurement. A tiny batch reactor is the solution, but there are three key problems: How to seal it; how to track catalytic turnover inside it; and how to see the nanoparticle inside it? Graphene solves all three problems: A microfabricated cavity with a thin SixNy bottom window, a single catalytic nanoparticle inside, and a graphene seal forms a gas tight batch reactor since graphene has zero gas permeability. Catalysis is then tracked as an internal pressure change via the stress & deflection of the graphene seal. Crucially, the electron-transparency of graphene and SixNy enables subsequent transmission electron microscope access with atomic resolution so that active nanoparticles can be studied in full detail. ATOMICAR will re-define the experimental limits of catalyst benchmarking and lift the field of basic catalysis research into the single-nanoparticle age.

1 496 000 €

Start date: 2018-02-01, End date: 2023-01-31

Deep learning is revolutionizing the field of Natural Language Processing (NLP), with breakthroughs in machine translation, speech recognition, and question answering. New language interfaces (digital assistants, messenger apps, customer service bots) are emerging as the next technologies for seamless, multilingual communication among humans and machines. From a machine learning perspective, many problems in NLP can be characterized as structured prediction: they involve predicting structurally rich and interdependent output variables. In spite of this, current neural NLP systems ignore the structural complexity of human language, relying on simplistic and error-prone greedy search procedures. This leads to serious mistakes in machine translation, such as words being dropped or named entities mistranslated. More broadly, neural networks are missing the key structural mechanisms for solving complex real-world tasks requiring deep reasoning. This project attacks these fundamental problems by bringing together deep learning and structured prediction, with a highly disruptive and cross-disciplinary approach. First, I will endow neural networks with a "planning mechanism" to guide structural search, letting decoders learn the optimal order by which they should operate. This makes a bridge with reinforcement learning and combinatorial optimization. Second, I will develop new ways of automatically inducing latent structure inside the network, making it more expressive, scalable and interpretable. Synergies with probabilistic inference and sparse modeling techniques will be exploited. To complement these two innovations, I will investigate new ways of incorporating weak supervision to reduce the need for labeled data. Three highly challenging applications will serve as testbeds: machine translation, quality estimation, and dependency parsing. To maximize technological impact, a collaboration is planned with a start-up company in the crowd-sourcing translation industry.

1 436 000 €

Start date: 2018-02-01, End date: 2023-01-31

Known, but mostly novel natural products (NPs) are in high demand – these are used in drugs, cosmetics and agrochemicals and serve also as research tools to probe biological systems. NP structures inspire chemists to develop new syntheses, and NP biosynthetic enzymes add to the metabolic engineer’s toolbox. The advent of next generation DNA-sequencing has revealed a vastly rich pool of NP biosynthetic gene clusters (BGCs) among bacterial genomes, most of which with no corresponding NP. Hence, opportunities abound for the discovery of new chemistry and enzymology that has the potential to push the boundaries of chemical space and enzymatic reactivity. Still, we cannot reliably predict chemistry from BGCs with unusual organization or encoding unknown functionalities, and, for molecules of unorthodox architecture, it is difficult to anticipate how their BGCs are organized. It is the valuable, truly novel chemistry and biochemistry that lies on these unexplored connections, that we aim to reveal with this proposal. To achieve it, we will work with a chemically-talented group of organisms – cyanobacteria, and with a specific structural class – fatty acids (FAs) – that is metabolized in a quite peculiar fashion by these organisms, paving the way for NP and enzyme discovery. On one hand, we will exploit the unique FA metabolism of cyanobacteria to develop a feeding strategy that will quickly reveal unprecedented FA-incorporating NPs. On the other, we will scrutinize the intriguing biosynthesis of three unique classes of metabolites that we have isolated recently and that incorporate and modify FA-moieties. We will find the BGCs for these compounds and dissect the functionality involved in such puzzling modifications to uncover important underlying enzymatic chemistry. This proposal is a blend of discovery- and hypothesis-driven research at the NP chemistry/biosynthesis interface that draws on the experience of the PI’s work on different aspects of cyanobacterial NPs.

1 462 938 €

Start date: 2018-01-01, End date: 2022-12-31

Society generates increasing amounts of data, which is both a resource and a challenge. The data reveal new insights that may potentially improve our livelihood, but their quantity renders such insights difficult to find. Machine learning techniques sift through the data looking for statistical patterns of interest to a given task. Due to an exponential growth in available data, these techniques enable us to automate difficult decisions, such as those needed for personalized medicine and self-driving cars. NoTape note that machine learning techniques depend on a distance measure to determine which data points are similar and which are not. As this measure is difficult to choose, NoTape develop methods for estimating an optimal distance measure directly from data. Empirical evidence suggest that the optimal distance measure in one region of data space need not coincide with the optimal measure in another region, i.e.that the distance measure should locally adapt to the data. Local adaptability imply that the distance measure itself will be sensitive to noise in the data, and therefore should be described as a random variable. NoTape estimate distance measures as random Riemannian metrics and perform statistical data analysis accordingly. The notion of statistical computations with respect to an uncertain locally adaptive distance measure is uncharted territory, which need new algorithms for numerical integration and for solving differential equations. As a guiding example, we estimate statistical models that reflect human perception. As perception processes are not fully understood, an optimal distance measure cannot be precisely estimated and the uncertainty of NoTape is needed. The geometric nature of the developed methods ensure that attained models are interpretable by humans, which contrast current locally adaptive techniques. As society automate more decisions, interpretability is increasing important to ensure that the machine learning system can be trusted.

1 463 805 €

Start date: 2017-12-01, End date: 2022-11-30