@unpublished{14601,
  abstract     = {In this work, we address the problem of learning provably stable neural
network policies for stochastic control systems. While recent work has
demonstrated the feasibility of certifying given policies using martingale
theory, the problem of how to learn such policies is little explored. Here, we
study the effectiveness of jointly learning a policy together with a martingale
certificate that proves its stability using a single learning algorithm. We
observe that the joint optimization problem becomes easily stuck in local
minima when starting from a randomly initialized policy. Our results suggest
that some form of pre-training of the policy is required for the joint
optimization to repair and verify the policy successfully.},
  author       = {Zikelic, Dorde and Lechner, Mathias and Chatterjee, Krishnendu and Henzinger, Thomas A},
  booktitle    = {arXiv},
  title        = {{Learning stabilizing policies in stochastic control systems}},
  doi          = {10.48550/arXiv.2205.11991},
  year         = {2022},
}

@article{11411,
  abstract     = {Many studies have quantified the distribution of heterozygosity and relatedness in natural populations, but few have examined the demographic processes driving these patterns. In this study, we take a novel approach by studying how population structure affects both pairwise identity and the distribution of heterozygosity in a natural population of the self-incompatible plant Antirrhinum majus. Excess variance in heterozygosity between individuals is due to identity disequilibrium, which reflects the variance in inbreeding between individuals; it is measured by the statistic g2. We calculated g2 together with FST and pairwise relatedness (Fij) using 91 SNPs in 22,353 individuals collected over 11 years. We find that pairwise Fij declines rapidly over short spatial scales, and the excess variance in heterozygosity between individuals reflects significant variation in inbreeding. Additionally, we detect an excess of individuals with around half the average heterozygosity, indicating either selfing or matings between close relatives. We use 2 types of simulation to ask whether variation in heterozygosity is consistent with fine-scale spatial population structure. First, by simulating offspring using parents drawn from a range of spatial scales, we show that the known pollen dispersal kernel explains g2. Second, we simulate a 1,000-generation pedigree using the known dispersal and spatial distribution and find that the resulting g2 is consistent with that observed from the field data. In contrast, a simulated population with uniform density underestimates g2, indicating that heterogeneous density promotes identity disequilibrium. Our study shows that heterogeneous density and leptokurtic dispersal can together explain the distribution of heterozygosity.},
  author       = {Surendranadh, Parvathy and Arathoon, Louise S and Baskett, Carina and Field, David and Pickup, Melinda and Barton, Nicholas H},
  issn         = {1943-2631},
  journal      = {Genetics},
  number       = {3},
  publisher    = {Oxford University Press},
  title        = {{Effects of fine-scale population structure on the distribution of heterozygosity in a long-term study of Antirrhinum majus}},
  doi          = {10.1093/genetics/iyac083},
  volume       = {221},
  year         = {2022},
}

@article{11842,
  abstract     = {We consider the flow of two viscous and incompressible fluids within a bounded domain modeled by means of a two-phase Navier–Stokes system. The two fluids are assumed to be immiscible, meaning that they are separated by an interface. With respect to the motion of the interface, we consider pure transport by the fluid flow. Along the boundary of the domain, a complete slip boundary condition for the fluid velocities and a constant ninety degree contact angle condition for the interface are assumed. In the present work, we devise for the resulting evolution problem a suitable weak solution concept based on the framework of varifolds and establish as the main result a weak-strong uniqueness principle in 2D. The proof is based on a relative entropy argument and requires a non-trivial further development of ideas from the recent work of Fischer and the first author (Arch. Ration. Mech. Anal. 236, 2020) to incorporate the contact angle condition. To focus on the effects of the necessarily singular geometry of the evolving fluid domains, we work for simplicity in the regime of same viscosities for the two fluids.},
  author       = {Hensel, Sebastian and Marveggio, Alice},
  issn         = {1422-6952},
  journal      = {Journal of Mathematical Fluid Mechanics},
  number       = {3},
  publisher    = {Springer Nature},
  title        = {{Weak-strong uniqueness for the Navier–Stokes equation for two fluids with ninety degree contact angle and same viscosities}},
  doi          = {10.1007/s00021-022-00722-2},
  volume       = {24},
  year         = {2022},
}

@unpublished{14597,
  abstract     = {Phase-field models such as the Allen-Cahn equation may give rise to the formation and evolution of geometric shapes, a phenomenon that may be analyzed rigorously in suitable scaling regimes. In its sharp-interface limit, the vectorial Allen-Cahn equation with a potential with N≥3 distinct minima has been conjectured to describe the evolution of branched interfaces by multiphase mean curvature flow.
In the present work, we give a rigorous proof for this statement in two and three ambient dimensions and for a suitable class of potentials: As long as a strong solution to multiphase mean curvature flow exists, solutions to the vectorial Allen-Cahn equation with well-prepared initial data converge towards multiphase mean curvature flow in the limit of vanishing interface width parameter ε↘0. We even establish the rate of convergence O(ε1/2).
Our approach is based on the gradient flow structure of the Allen-Cahn equation and its limiting motion: Building on the recent concept of "gradient flow calibrations" for multiphase mean curvature flow, we introduce a notion of relative entropy for the vectorial Allen-Cahn equation with multi-well potential. This enables us to overcome the limitations of other approaches, e.g. avoiding the need for a stability analysis of the Allen-Cahn operator or additional convergence hypotheses for the energy at positive times.},
  author       = {Fischer, Julian L and Marveggio, Alice},
  booktitle    = {arXiv},
  title        = {{Quantitative convergence of the vectorial Allen-Cahn equation towards multiphase mean curvature flow}},
  doi          = {10.48550/ARXIV.2203.17143},
  year         = {2022},
}

@article{10791,
  abstract     = {The mammalian neocortex is composed of diverse neuronal and glial cell classes that broadly arrange in six distinct laminae. Cortical layers emerge during development and defects in the developmental programs that orchestrate cortical lamination are associated with neurodevelopmental diseases. The developmental principle of cortical layer formation depends on concerted radial projection neuron migration, from their birthplace to their final target position. Radial migration occurs in defined sequential steps, regulated by a large array of signaling pathways. However, based on genetic loss-of-function experiments, most studies have thus far focused on the role of cell-autonomous gene function. Yet, cortical neuron migration in situ is a complex process and migrating neurons traverse along diverse cellular compartments and environments. The role of tissue-wide properties and genetic state in radial neuron migration is however not clear. Here we utilized mosaic analysis with double markers (MADM) technology to either sparsely or globally delete gene function, followed by quantitative single-cell phenotyping. The MADM-based gene ablation paradigms in combination with computational modeling demonstrated that global tissue-wide effects predominate cell-autonomous gene function albeit in a gene-specific manner. Our results thus suggest that the genetic landscape in a tissue critically affects the overall migration phenotype of individual cortical projection neurons. In a broader context, our findings imply that global tissue-wide effects represent an essential component of the underlying etiology associated with focal malformations of cortical development in particular, and neurological diseases in general.},
  author       = {Hansen, Andi H and Pauler, Florian and Riedl, Michael and Streicher, Carmen and Heger, Anna-Magdalena and Laukoter, Susanne and Sommer, Christoph M and Nicolas, Armel and Hof, Björn and Tsai, Li Huei and Rülicke, Thomas and Hippenmeyer, Simon},
  issn         = {2753-149X},
  journal      = {Oxford Open Neuroscience},
  number       = {1},
  publisher    = {Oxford University Press},
  title        = {{Tissue-wide effects override cell-intrinsic gene function in radial neuron migration}},
  doi          = {10.1093/oons/kvac009},
  volume       = {1},
  year         = {2022},
}

@inproceedings{12299,
  abstract     = {Transfer learning is a classic paradigm by which models pretrained on large “upstream” datasets are adapted to yield good results on “downstream” specialized datasets. Generally, more accurate models on the “upstream” dataset tend to provide better transfer accuracy “downstream”. In this work, we perform an in-depth investigation of this phenomenon in the context of convolutional neural networks (CNNs) trained on the ImageNet dataset, which have been pruned-that is, compressed by sparsifiying their connections. We consider transfer using unstructured pruned models obtained by applying several state-of-the-art pruning methods, including magnitude-based, second-order, regrowth, lottery-ticket, and regularization approaches, in the context of twelve standard transfer tasks. In a nutshell, our study shows that sparse models can match or even outperform the transfer performance of dense models, even at high sparsities, and, while doing so, can lead to significant inference and even training speedups. At the same time, we observe and analyze significant differences in the behaviour of different pruning methods. The code is available at: https://github.com/IST-DASLab/sparse-imagenet-transfer.},
  author       = {Iofinova, Eugenia B and Peste, Elena-Alexandra and Kurtz, Mark and Alistarh, Dan-Adrian},
  booktitle    = {2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  issn         = {2575-7075},
  location     = {New Orleans, LA, United States},
  pages        = {12256--12266},
  publisher    = {Institute of Electrical and Electronics Engineers},
  title        = {{How well do sparse ImageNet models transfer?}},
  doi          = {10.1109/cvpr52688.2022.01195},
  year         = {2022},
}

@phdthesis{11128,
  abstract     = {Although we often see studies focusing on simple or even discrete traits in studies of colouration,
the variation of “appearance” phenotypes found in nature is often more complex, continuous
and high-dimensional. Therefore, we developed automated methods suitable for large datasets
of genomes and images, striving to account for their complex nature, while minimising human
bias. We used these methods on a dataset of more than 20, 000 plant SNP genomes and
corresponding fower images from a hybrid zone of two subspecies of Antirrhinum majus with
distinctly coloured fowers to improve our understanding of the genetic nature of the fower
colour in our study system.
Firstly, we use the advantage of large numbers of genotyped plants to estimate the haplotypes in
the main fower colour regulating region. We study colour- and geography-related characteristics
of the estimated haplotypes and how they connect to their relatedness. We show discrepancies
from the expected fower colour distributions given the genotype and identify particular
haplotypes leading to unexpected phenotypes. We also confrm a signifcant defcit of the
double recessive recombinant and quite surprisingly, we show that haplotypes of the most
frequent parental type are much less variable than others.
Secondly, we introduce our pipeline capable of processing tens of thousands of full fower
images without human interaction and summarising each image into a set of informative scores.
We show the compatibility of these machine-measured fower colour scores with the previously
used manual scores and study impact of external efect on the resulting scores. Finally, we use
the machine-measured fower colour scores to ft and examine a phenotype cline across the
hybrid zone in Planoles using full fower images as opposed to discrete, manual scores and
compare it with the genotypic cline.},
  author       = {Matejovicova, Lenka},
  isbn         = {978-3-99078-016-9},
  issn         = {2663-337X},
  pages        = {112},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{Genetic basis of flower colour as a model for adaptive evolution}},
  doi          = {10.15479/at:ista:11128},
  year         = {2022},
}

@phdthesis{12072,
  abstract     = {In this thesis, we study two of the most important questions in Arithmetic geometry: that of the existence and density of solutions to Diophantine equations. In order for a Diophantine equation to have any solutions over the rational numbers, it must have solutions everywhere locally, i.e., over R and over Qp for every prime p. The converse, called the Hasse principle, is known to fail in general. However, it is still a central question in Arithmetic geometry to determine for which varieties the Hasse principle does hold. In this work, we establish the Hasse principle for a wide new family of varieties of the form f(t) = NK/Q(x) ̸= 0, where f is a polynomial with integer coefficients and NK/Q denotes the norm
form associated to a number field K. Our results cover products of arbitrarily many linear, quadratic or cubic factors, and generalise an argument of Irving [69], which makes use of the beta sieve of Rosser and Iwaniec. We also demonstrate how our main sieve results can be applied to treat new cases of a conjecture of Harpaz and Wittenberg on locally split values of polynomials over number fields, and discuss consequences for rational points in fibrations.
In the second question, about the density of solutions, one defines a height function and seeks to estimate asymptotically the number of points of height bounded by B as B → ∞. Traditionally, one either counts rational points, or
integral points with respect to a suitable model. However, in this thesis, we study an emerging area of interest in Arithmetic geometry known as Campana points, which in some sense interpolate between rational and integral points.
More precisely, we count the number of nonzero integers z1, z2, z3 such that gcd(z1, z2, z3) = 1, and z1, z2, z3, z1 + z2 + z3 are all squareful and bounded by B. Using the circle method, we obtain an asymptotic formula which agrees in
the power of B and log B with a bold new generalisation of Manin’s conjecture to the setting of Campana points, recently formulated by Pieropan, Smeets, Tanimoto and Várilly-Alvarado [96]. However, in this thesis we also provide the first known counterexamples to leading constant predicted by their conjecture. },
  author       = {Shute, Alec L},
  isbn         = {978-3-99078-023-7},
  issn         = {2663-337X},
  pages        = {208},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{Existence and density problems in Diophantine geometry: From norm forms to Campana points}},
  doi          = {10.15479/at:ista:12072},
  year         = {2022},
}

@phdthesis{11777,
  abstract     = {In this dissertation we study coboundary expansion of simplicial complex with a view of giving geometric applications.
Our main novel tool is an equivariant version of Gromov's celebrated Topological Overlap Theorem. The equivariant topological overlap theorem leads to various geometric applications including a quantitative non-embeddability result for sufficiently thick buildings (which partially resolves a conjecture of Tancer and Vorwerk) and an improved lower bound on the pair-crossing number of (bounded degree) expander graphs. Additionally, we will give new proofs for several known lower bounds for geometric problems such as the number of Tverberg partitions or the crossing number of complete bipartite graphs.
For the aforementioned applications one is naturally lead to study expansion properties of joins of simplicial complexes. In the presence of a special certificate for expansion (as it is the case, e.g., for spherical buildings), the join of two expanders is an expander. On the flip-side, we report quite some evidence that coboundary expansion exhibits very non-product-like behaviour under taking joins. For instance, we exhibit infinite families of graphs $(G_n)_{n\in \mathbb{N}}$ and $(H_n)_{n\in\mathbb{N}}$ whose join $G_n*H_n$ has expansion of lower order than the product of the expansion constant of the graphs. Moreover, we show an upper bound of $(d+1)/2^d$ on the normalized coboundary expansion constants for the complete multipartite complex $[n]^{*(d+1)}$ (under a mild divisibility condition on $n$).
Via the probabilistic method the latter result extends to an upper bound of $(d+1)/2^d+\varepsilon$ on the coboundary expansion constant of the spherical building associated with $\mathrm{PGL}_{d+2}(\mathbb{F}_q)$ for any $\varepsilon>0$ and sufficiently large $q=q(\varepsilon)$. This disproves a conjecture of Lubotzky, Meshulam and Mozes -- in a rather strong sense.
By improving on existing lower bounds we make further progress towards closing the gap between the known lower and upper bounds on the coboundary expansion constants of $[n]^{*(d+1)}$. The best improvements we achieve using computer-aided proofs and flag algebras. The exact value even for the complete $3$-partite $2$-dimensional complex $[n]^{*3}$ remains unknown but we are happy to conjecture a precise value for every $n$. %Moreover, we show that a previously shown lower bound on the expansion constant of the spherical building associated with $\mathrm{PGL}_{2}(\mathbb{F}_q)$ is not tight.
In a loosely structured, last chapter of this thesis we collect further smaller observations related to expansion. We point out a link between discrete Morse theory and a technique for showing coboundary expansion, elaborate a bit on the hardness of computing coboundary expansion constants, propose a new criterion for coboundary expansion (in a very dense setting) and give one way of making the folklore result that expansion of links is a necessary condition for a simplicial complex to be an expander precise.},
  author       = {Wild, Pascal},
  isbn         = {978-3-99078-021-3},
  issn         = {2663-337X},
  pages        = {170},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{High-dimensional expansion and crossing numbers of simplicial complexes}},
  doi          = {10.15479/at:ista:11777},
  year         = {2022},
}

@phdthesis{11473,
  abstract     = {The polaron model is a basic model of quantum field theory describing a single particle
interacting with a bosonic field. It arises in many physical contexts. We are mostly concerned
with models applicable in the context of an impurity atom in a Bose-Einstein condensate as
well as the problem of electrons moving in polar crystals.
The model has a simple structure in which the interaction of the particle with the field is given
by a term linear in the field’s creation and annihilation operators. In this work, we investigate
the properties of this model by providing rigorous estimates on various energies relevant to the
problem. The estimates are obtained, for the most part, by suitable operator techniques which
constitute the principal mathematical substance of the thesis.
The first application of these techniques is to derive the polaron model rigorously from first
principles, i.e., from a full microscopic quantum-mechanical many-body problem involving an
impurity in an otherwise homogeneous system. We accomplish this for the N + 1 Bose gas
in the mean-field regime by showing that a suitable polaron-type Hamiltonian arises at weak
interactions as a low-energy effective theory for this problem.
In the second part, we investigate rigorously the ground state of the model at fixed momentum
and for large values of the coupling constant. Qualitatively, the system is expected to display
a transition from the quasi-particle behavior at small momenta, where the dispersion relation
is parabolic and the particle moves through the medium dragging along a cloud of phonons, to
the radiative behavior at larger momenta where the polaron decelerates and emits free phonons.
At the same time, in the strong coupling regime, the bosonic field is expected to behave purely
classically. Accordingly, the effective mass of the polaron at strong coupling is conjectured to
be asymptotically equal to the one obtained from the semiclassical counterpart of the problem,
first studied by Landau and Pekar in the 1940s. For polaron models with regularized form
factors and phonon dispersion relations of superfluid type, i.e., bounded below by a linear
function of the wavenumbers for all phonon momenta as in the interacting Bose gas, we prove
that for a large window of momenta below the radiation threshold, the energy-momentum
relation at strong coupling is indeed essentially a parabola with semi-latus rectum equal to the
Landau–Pekar effective mass, as expected.
For the Fröhlich polaron describing electrons in polar crystals where the dispersion relation is
of the optical type and the form factor is formally UV–singular due to the nature of the point
charge-dipole interaction, we are able to give the corresponding upper bound. In contrast to
the regular case, this requires the inclusion of the quantum fluctuations of the phonon field,
which makes the problem considerably more difficult.
The results are supplemented by studies on the absolute ground-state energy at strong coupling,
a proof of the divergence of the effective mass with the coupling constant for a wide class of
polaron models, as well as the discussion of the apparent UV singularity of the Fröhlich model
and the application of the techniques used for its removal for the energy estimates.
},
  author       = {Mysliwy, Krzysztof},
  issn         = {2663-337X},
  pages        = {138},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{Polarons in Bose gases and polar crystals: Some rigorous energy estimates}},
  doi          = {10.15479/at:ista:11473},
  year         = {2022},
}

@article{11995,
  abstract     = {G protein-coupled receptors (GPCRs) regulate processes ranging from immune responses to neuronal signaling. However, ligands for many GPCRs remain unknown, suffer from off-target effects or have poor bioavailability. Additionally, dissecting cell type-specific responses is challenging when the same GPCR is expressed on different cells within a tissue. Here, we overcome these limitations by engineering DREADD-based GPCR chimeras that bind clozapine-N-oxide and mimic a GPCR-of-interest. We show that chimeric DREADD-β2AR triggers responses comparable to β2AR on second messenger and kinase activity, post-translational modifications, and protein-protein interactions. Moreover, we successfully recapitulate β2AR-mediated filopodia formation in microglia, an immune cell capable of driving central nervous system inflammation. When dissecting microglial inflammation, we included two additional DREADD-based chimeras mimicking microglia-enriched GPR65 and GPR109A. DREADD-β2AR and DREADD-GPR65 modulate the inflammatory response with high similarity to endogenous β2AR, while DREADD-GPR109A shows no impact. Our DREADD-based approach allows investigation of cell type-dependent pathways without known endogenous ligands.},
  author       = {Schulz, Rouven and Korkut, Medina and Venturino, Alessandro and Colombo, Gloria and Siegert, Sandra},
  issn         = {2041-1723},
  journal      = {Nature Communications},
  publisher    = {Springer Nature},
  title        = {{Chimeric GPCRs mimic distinct signaling pathways and modulate microglia responses}},
  doi          = {10.1038/s41467-022-32390-1},
  volume       = {13},
  year         = {2022},
}

@article{10564,
  abstract     = {We study a class of polaron-type Hamiltonians with sufficiently regular form factor in the interaction term. We investigate the strong-coupling limit of the model, and prove suitable bounds on the ground state energy as a function of the total momentum of the system. These bounds agree with the semiclassical approximation to leading order. The latter corresponds here to the situation when the particle undergoes harmonic motion in a potential well whose frequency is determined by the corresponding Pekar functional. We show that for all such models the effective mass diverges in the strong coupling limit, in all spatial dimensions. Moreover, for the case when the phonon dispersion relation grows at least linearly with momentum, the bounds result in an asymptotic formula for the effective mass quotient, a quantity generalizing the usual notion of the effective mass. This asymptotic form agrees with the semiclassical Landau–Pekar formula and can be regarded as the first rigorous confirmation, in a slightly weaker sense than usually considered, of the validity of the semiclassical formula for the effective mass.},
  author       = {Mysliwy, Krzysztof and Seiringer, Robert},
  issn         = {1572-9613},
  journal      = {Journal of Statistical Physics},
  number       = {1},
  publisher    = {Springer Nature},
  title        = {{Polaron models with regular interactions at strong coupling}},
  doi          = {10.1007/s10955-021-02851-w},
  volume       = {186},
  year         = {2022},
}

@phdthesis{11945,
  abstract     = {G protein-coupled receptors (GPCRs) respond to specific ligands and regulate multiple processes ranging from cell growth and immune responses to neuronal signal transmission. However, ligands for many GPCRs remain unknown, suffer from off-target effects or have poor bioavailability. Additional challenges exist to dissect cell-type specific responses when the same GPCR is expressed on several cell types within the body. Here, we overcome these limitations by engineering DREADD-based GPCR chimeras that selectively bind their agonist clozapine-N-oxide (CNO) and mimic a GPCR-of-interest in a desired cell type.
We validated our approach with β2-adrenergic receptor (β2AR/ADRB2) and show that our chimeric DREADD-β2AR triggers comparable responses on second messenger and kinase activity, post-translational modifications, and protein-protein interactions. Since β2AR is also enriched in microglia, which can drive inflammation in the central nervous system, we expressed chimeric DREADD-β2AR in primary microglia and successfully recapitulate β2AR-mediated filopodia formation through CNO stimulation. To dissect the role of selected GPCRs during microglial inflammation, we additionally generated DREADD-based chimeras for microglia-enriched GPR65 and GPR109A/HCAR2. In a microglia cell line, DREADD-β2AR and DREADD-GPR65 both modulated the inflammatory response with a similar profile as endogenously expressed β2AR, while DREADD-GPR109A showed no impact.
Our DREADD-based approach provides the means to obtain mechanistic and functional insights into GPCR signaling on a cell-type specific level.},
  author       = {Schulz, Rouven},
  issn         = {2663-337X},
  pages        = {133},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{Chimeric G protein-coupled receptors mimic distinct signaling pathways and modulate microglia function}},
  doi          = {10.15479/at:ista:11945},
  year         = {2022},
}

@article{10411,
  abstract     = {The phytohormone auxin is the major growth regulator governing tropic responses including gravitropism. Auxin build-up at the lower side of stimulated shoots promotes cell expansion, whereas in roots it inhibits growth, leading to upward shoot bending and downward root bending, respectively. Yet it remains an enigma how the same signal can trigger such opposite cellular responses. In this review, we discuss several recent unexpected insights into the mechanisms underlying auxin regulation of growth, challenging several existing models. We focus on the divergent mechanisms of apoplastic pH regulation in shoots and roots revisiting the classical Acid Growth Theory and discuss coordinated involvement of multiple auxin signaling pathways. From this emerges a more comprehensive, updated picture how auxin regulates growth.},
  author       = {Li, Lanxin and Gallei, Michelle C and Friml, Jiří},
  issn         = {1360-1385},
  journal      = {Trends in Plant Science},
  number       = {5},
  pages        = {440--449},
  publisher    = {Cell Press},
  title        = {{Bending to auxin: Fast acid growth for tropisms}},
  doi          = {10.1016/j.tplants.2021.11.006},
  volume       = {27},
  year         = {2022},
}

@phdthesis{10799,
  abstract     = {Because of the increasing popularity of machine learning methods, it is becoming important to understand the impact of learned components on automated decision-making systems and to guarantee that their consequences are beneficial to society. In other words, it is necessary to ensure that machine learning is sufficiently trustworthy to be used in real-world applications. This thesis studies two properties of machine learning models that are highly desirable for the
sake of reliability: robustness and fairness. In the first part of the thesis we study the robustness of learning algorithms to training data corruption. Previous work has shown that machine learning models are vulnerable to a range
of training set issues, varying from label noise through systematic biases to worst-case data manipulations. This is an especially relevant problem from a present perspective, since modern machine learning methods are particularly data hungry and therefore practitioners often have to rely on data collected from various external sources, e.g. from the Internet, from app users or via crowdsourcing. Naturally, such sources vary greatly in the quality and reliability of the
data they provide. With these considerations in mind, we study the problem of designing machine learning algorithms that are robust to corruptions in data coming from multiple sources. We show that, in contrast to the case of a single dataset with outliers, successful learning within this model is possible both theoretically and practically, even under worst-case data corruptions. The second part of this thesis deals with fairness-aware machine learning. There are multiple areas where machine learning models have shown promising results, but where careful considerations are required, in order to avoid discrimanative decisions taken by such learned components. Ensuring fairness can be particularly challenging, because real-world training datasets are expected to contain various forms of historical bias that may affect the learning process. In this thesis we show that data corruption can indeed render the problem of achieving fairness impossible, by tightly characterizing the theoretical limits of fair learning under worst-case data manipulations. However, assuming access to clean data, we also show how fairness-aware learning can be made practical in contexts beyond binary classification, in particular in the challenging learning to rank setting.},
  author       = {Konstantinov, Nikola H},
  isbn         = {978-3-99078-015-2},
  issn         = {2663-337X},
  keywords     = {robustness, fairness, machine learning, PAC learning, adversarial learning},
  pages        = {176},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{Robustness and fairness in machine learning}},
  doi          = {10.15479/at:ista:10799},
  year         = {2022},
}

@article{10802,
  abstract     = {Addressing fairness concerns about machine learning models is a crucial step towards their long-term adoption in real-world automated systems. While many approaches have been developed for training fair models from data, little is known about the robustness of these methods to data corruption. In this work we consider fairness-aware learning under worst-case data manipulations. We show that an adversary can in some situations force any learner to return an overly biased classifier, regardless of the sample size and with or without degrading
accuracy, and that the strength of the excess bias increases for learning problems with underrepresented protected groups in the data. We also prove that our hardness results are tight up to constant factors. To this end, we study two natural learning algorithms that optimize for both accuracy and fairness and show that these algorithms enjoy guarantees that are order-optimal in terms of the corruption ratio and the protected groups frequencies in the large data
limit.},
  author       = {Konstantinov, Nikola H and Lampert, Christoph},
  issn         = {1533-7928},
  journal      = {Journal of Machine Learning Research},
  keywords     = {Fairness, robustness, data poisoning, trustworthy machine learning, PAC learning},
  pages        = {1--60},
  publisher    = {ML Research Press},
  title        = {{Fairness-aware PAC learning from corrupted data}},
  volume       = {23},
  year         = {2022},
}

@unpublished{11366,
  abstract     = {Adversarial training (i.e., training on adversarially perturbed input data) is a well-studied method for making neural networks robust to potential adversarial attacks during inference. However, the improved robustness does not
come for free but rather is accompanied by a decrease in overall model accuracy and performance. Recent work has shown that, in practical robot learning applications, the effects of adversarial training do not pose a fair trade-off
but inflict a net loss when measured in holistic robot performance. This work revisits the robustness-accuracy trade-off in robot learning by systematically analyzing if recent advances in robust training methods and theory in
conjunction with adversarial robot learning can make adversarial training suitable for real-world robot applications. We evaluate a wide variety of robot learning tasks ranging from autonomous driving in a high-fidelity environment
amenable to sim-to-real deployment, to mobile robot gesture recognition. Our results demonstrate that, while these techniques make incremental improvements on the trade-off on a relative scale, the negative side-effects caused by
adversarial training still outweigh the improvements by an order of magnitude. We conclude that more substantial advances in robust learning methods are necessary before they can benefit robot learning tasks in practice.},
  author       = {Lechner, Mathias and Amini, Alexander and Rus, Daniela and Henzinger, Thomas A},
  booktitle    = {arXiv},
  title        = {{Revisiting the adversarial robustness-accuracy tradeoff in robot learning}},
  doi          = {10.48550/arXiv.2204.07373},
  year         = {2022},
}

@unpublished{21673,
  abstract     = {When impinging on optical structures or passing in their vicinity, free electrons can spontaneously emit electromagnetic radiation, a phenomenon generally known as cathodoluminescence. Free-electron radiation comes in many guises: Cherenkov, transition, and Smith-Purcell radiation, but also electron scintillation, commonly referred to as incoherent cathodoluminescence. While those effects have been at the heart of many fundamental discoveries and technological developments in high-energy physics in the past century, their recent demonstration in photonic and nanophotonic systems has attracted a lot of attention. Those developments arose from predictions that exploit nanophotonics for novel radiation regimes, now becoming accessible thanks to advances in nanofabrication. In general, the proper design of nanophotonic structures can enable shaping, control, and enhancement of free-electron radiation, for any of the above-mentioned effects. Free-electron radiation in nanophotonics opens the way to promising applications, such as widely-tunable integrated light sources from x-ray to THz frequencies, miniaturized particle accelerators, and highly sensitive high-energy particle detectors. Here, we review the emerging field of free-electron radiation in nanophotonics. We first present a general, unified framework to describe free-electron light-matter interaction in arbitrary nanophotonic systems. We then show how this framework sheds light on the physical underpinnings of many methods in the field used to control and enhance free-electron radiation. Namely, the framework points to the central role played by the photonic eigenmodes in controlling the output properties of free-electron radiation (e.g., frequency, directionality, and polarization). [... see full abstract in paper]},
  author       = {Roques-Carmes, Charles and Kooi, Steven E. and Yang, Yi and Rivera, Nicholas and Keathley, Phillip D. and Joannopoulos, John D. and Johnson, Steven G. and Kaminer, Ido and Berggren, Karl K. and Soljačić, Marin},
  booktitle    = {arXiv},
  title        = {{Free-electron-light interactions in nanophotonics}},
  doi          = {10.48550/arXiv.2208.02368},
  year         = {2022},
}

@article{11344,
  abstract     = {Until recently, Shigella and enteroinvasive Escherichia coli were thought to be primate-restricted pathogens. The base of their pathogenicity is the type 3 secretion system (T3SS) encoded by the pINV virulence plasmid, which facilitates host cell invasion and subsequent proliferation. A large family of T3SS effectors, E3 ubiquitin-ligases encoded by the ipaH genes, have a key role in the Shigella pathogenicity through the modulation of cellular ubiquitination that degrades host proteins. However, recent genomic studies identified ipaH genes in the genomes of Escherichia marmotae, a potential marmot pathogen, and an E. coli extracted from fecal samples of bovine calves, suggesting that non-human hosts may also be infected by these strains, potentially pathogenic to humans. We performed a comparative genomic study of the functional repertoires in the ipaH gene family in Shigella and enteroinvasive Escherichia from human and predicted non-human hosts. We found that fewer than half of Shigella genomes had a complete set of ipaH genes, with frequent gene losses and duplications that were not consistent with the species tree and nomenclature. Non-human host IpaH proteins had a diverse set of substrate-binding domains and, in contrast to the Shigella proteins, two variants of the NEL C-terminal domain. Inconsistencies between strains phylogeny and composition of effectors indicate horizontal gene transfer between E. coli adapted to different hosts. These results provide a framework for understanding of ipaH-mediated host-pathogens interactions and suggest a need for a genomic study of fecal samples from diseased animals.},
  author       = {Dranenko, NO and Tutukina, MN and Gelfand, MS and Kondrashov, Fyodor and Bochkareva, Olga},
  issn         = {2045-2322},
  journal      = {Scientific Reports},
  publisher    = {Springer Nature},
  title        = {{Chromosome-encoded IpaH ubiquitin ligases indicate non-human enteroinvasive Escherichia}},
  doi          = {10.1038/s41598-022-10827-3},
  volume       = {12},
  year         = {2022},
}

@phdthesis{12390,
  abstract     = {The scope of this thesis is to study quantum systems exhibiting a continuous symmetry that
is broken on the level of the corresponding effective theory. In particular we are going to
investigate translation-invariant Bose gases in the mean field limit, effectively described by
the Hartree functional, and the Fröhlich Polaron in the regime of strong coupling, effectively
described by the Pekar functional. The latter is a model describing the interaction between a
charged particle and the optical modes of a polar crystal. Regarding the former, we assume in
addition that the particles in the gas are unconfined, and typically we will consider particles
that are subject to an attractive interaction. In both cases the ground state energy of the
Hamiltonian is not a proper eigenvalue due to the underlying translation-invariance, while on
the contrary there exists a whole invariant orbit of minimizers for the corresponding effective
functionals. Both, the absence of proper eigenstates and the broken symmetry of the effective
theory, make the study significantly more involved and it is the content of this thesis to
develop a frameworks which allows for a systematic way to circumvent these issues.
It is a well-established result that the ground state energy of Bose gases in the mean field limit,
as well as the ground state energy of the Fröhlich Polaron in the regime of strong coupling, is
to leading order given by the minimal energy of the corresponding effective theory. As part
of this thesis we identify the sub-leading term in the expansion of the ground state energy,
which can be interpreted as the quantum correction to the classical energy, since the effective
theories under consideration can be seen as classical counterparts.
We are further going to establish an asymptotic expression for the energy-momentum relation
of the Fröhlich Polaron in the strong coupling limit. In the regime of suitably small momenta,
this asymptotic expression agrees with the energy-momentum relation of a free particle having
an effectively increased mass, and we find that this effectively increased mass agrees with the
conjectured value in the physics literature.
In addition we will discuss two unrelated papers written by the author during his stay at ISTA
in the appendix. The first one concerns the realization of anyons, which are quasi-particles
acquiring a non-trivial phase under the exchange of two particles, as molecular impurities.
The second one provides a classification of those vector fields defined on a given manifold
that can be written as the gradient of a given functional with respect to a suitable metric,
provided that some mild smoothness assumptions hold. This classification is subsequently
used to identify those quantum Markov semigroups that can be written as a gradient flow of
the relative entropy.
},
  author       = {Brooks, Morris},
  issn         = {2663-337X},
  pages        = {196},
  publisher    = {Institute of Science and Technology Austria},
  title        = {{Translation-invariant quantum systems with effectively broken symmetry}},
  doi          = {10.15479/at:ista:12390},
  year         = {2022},
}