Search

Building consensus sequences based on distributed, fault-tolerant consensus, as used for replicated state machines, typically requires a separate distributed state for every new consensus instance. Allocating and maintaining this state causes significant overhead. In particular, freeing the distributed, outdated states in a fault-tolerant way is not trivial and adds further complexity and cost to the system. In this paper, we propose an extension to the single-decree Paxos protocol that can learn a sequence of consensus decisions 'in-place', i.e. with a single set of distributed states. Our protocol does not require dynamic log structures and hence has no need for distributed log pruning, snapshotting, compaction, or dynamic resource allocation. The protocol builds a fault-tolerant atomic register that supports arbitrary read-modify-write operations. We use the concept of consistent quorums to detect whether the previous consensus still needs to be consolidated or is already finished so that the next consensus value can be safely proposed. Reading a consolidated consensus is done without state modifications and is thereby free of concurrency control and demand for serialisation. A proposer that is not interrupted reaches agreement on consecutive consensus decisions within a single message round-trip per decision by preparing the acceptors eagerly with the previous request.

Deep convolutional neural networks (CNNs) have shown outstanding performance in the task of semantically segmenting images. However, applying the same methods on 3D data still poses challenges due to the heavy memory requirements and the lack of structured data. Here, we propose LatticeNet, a novel approach for 3D semantic segmentation, which takes as input raw point clouds. A PointNet describes the local geometry which we embed into a sparse permutohedral lattice. The lattice allows for fast convolutions while keeping a low memory footprint. Further, we introduce DeformSlice, a novel learned data-dependent interpolation for projecting lattice features back onto the point cloud. We present results of 3D segmentation on various datasets where our method achieves state-of-the-art performance.

The convex hull of N independent random points chosen on the boundary of a simple polytope in R^n is investigated. Asymptotic formulas for the expected number of vertices and facets, and for the expectation of the volume difference are derived. This is one of the first investigations leading to rigorous results for random polytopes which are neither simple nor simplicial. The results contrast existing results when points are chosen in the interior of a convex set.

The GERmanium Detector Array (GERDA) experiment at the Gran Sasso underground laboratory (LNGS) of INFN is searching for neutrinoless double-beta ($0\nu\beta\beta$) decay of $^{76}$Ge. The technological challenge of GERDA is to operate in a "background-free" regime in the region of interest (ROI) after analysis cuts for the full 100$\,$kg$\cdot$yr target exposure of the experiment. A careful modeling and decomposition of the full-range energy spectrum is essential to predict the shape and composition of events in the ROI around $Q_{\beta\beta}$ for the $0\nu\beta\beta$ search, to extract a precise measurement of the half-life of the double-beta decay mode with neutrinos ($2\nu\beta\beta$) and in order to identify the location of residual impurities. The latter will permit future experiments to build strategies in order to further lower the background and achieve even better sensitivities. In this article the background decomposition prior to analysis cuts is presented for GERDA Phase II. The background model fit yields a flat spectrum in the ROI with a background index (BI) of $16.04^{+0.78}_{-0.85} \cdot 10^{-3}\,$cts/(kg$\cdot$keV$\cdot$yr) for the enriched BEGe data set and $14.68^{+0.47}_{-0.52} \cdot 10^{-3}\,$cts/(kg$\cdot$keV$\cdot$yr) for the enriched coaxial data set. These values are similar to the one of Gerda Phase I despite a much larger number of detectors and hence radioactive hardware components.

The paper argues that the material scope of AI regulations should not rely on the term "artificial intelligence (AI)". The argument is developed by proposing a number of requirements for legal definitions, surveying existing AI definitions, and then discussing the extent to which they meet the proposed requirements. It is shown that existing definitions of AI do not meet the most important requirements for legal definitions. Next, the paper argues that a risk-based approach would be preferable. Rather than using the term AI, policy makers should focus on the specific risks they want to reduce. It is shown that the requirements for legal definitions can be better met by defining the main sources of relevant risks: certain technical approaches (e.g. reinforcement learning), applications (e.g. facial recognition), and capabilities (e.g. the ability to physically interact with the environment). Finally, the paper discusses the extent to which this approach can also be applied to more advanced AI systems., Comment: Forthcoming in Law, Innovation and Technology, Volume 15, Issue 1. A previous version of this paper was titled "A legal definition of AI". The current version has been completely revised

Given a convex body K in R^n and p in R, we introduce and study the extremal inner and outer affine surface areas IS_p(K) = sup_{K'\subseteq K} (as_p(K') ) and os_p(K)=inf_{K'\supseteq K} (as_p(K') ), where as_p(K') denotes the L_p-affine surface area of K', and the supremum is taken over all convex subsets of K and the infimum over all convex compact subsets containing K. The convex body that realizes IS_1(K) in dimension 2 was determined by Barany. He also showed that this body is the limit shape of lattice polytopes in K. In higher dimensions no results are known about the extremal bodies. We use a thin shell estimate of Guedon and Milman and the L\"owner ellipsoid to give asymptotic estimates on the size of IS_p(K) and os_p(K). Surprisingly, both quantities are proportional to a power of volume.

Recent experimental progress in the fields of cold quantum gases and ultrafast optical spectroscopy of quantum materials allows to controllably induce and probe non-adiabatic dynamics of superconductors and superfluids. The time-evolution of the gap function before relaxation with the lattice is determined by the superposition of coherently evolving individual Cooper pairs within the manifold of the Bardeen-Cooper-Schrieffer (BCS) wavefunction. While dynamics following an abrupt quench of the pairing interaction strength in the single-band BCS model has been exactly solved due to the integrability of the model, the dynamics of post-quench multi-band superconductors remain under scrutiny. Here, we develop a generalization of the Volkov-Kogan Laplace-space perturbative method that allows us to determine the non-adiabatic gap dynamics of two-band fully gapped superconductors for a wide range of quench amplitudes. Our approach expands the long-time dynamics around the steady-state asymptotic value of the gap, which is self-consistently determined, rather than around the equilibrium value of the gap. We explicitly demonstrate that this method recovers the exact solution of the long-time gap dynamics in the single-band case and perfectly agrees with a numerical solution of the two-band model. We discover that dephasing of Cooper pairs from different bands leads to faster collisionless relaxation of the gap oscillation with a power-law of $t^{-3/2}$ instead of the well-known $t^{-1/2}$ behavior found in the single-band case. Furthermore, the gap oscillations display beating patterns arising from the existence of two different asymptotic gap values. Our results have important implications to a variety of two-band superconductors driven out of equilibrium, such as iron-based superconductors, MgB$_{2}$, and SrTiO$_{3}$., Comment: 18 pages, 9 figures

Given d in IN, we prove that all smooth K3 surfaces (over any field of characteristic p other than 2,3) of degree greater than 84d^2 contain at most 24 rational curves of degree at most d. In the exceptional characteristics, the same bounds hold for non-unirational K3 surfaces, and we develop analogous results in the unirational case. For d at least 3, we also construct K3 surfaces of any degree greater than 4d(d+1) with 24 rational curves of degree exactly d, thus attaining the above bounds., Comment: v3: 21 pages; final version with minor changes

Machine learning advances chemistry and materials science by enabling large-scale exploration of chemical space based on quantum chemical calculations. While these models supply fast and accurate predictions of atomistic chemical properties, they do not explicitly capture the electronic degrees of freedom of a molecule, which limits their applicability for reactive chemistry and chemical analysis. Here we present a deep learning framework for the prediction of the quantum mechanical wavefunction in a local basis of atomic orbitals from which all other ground-state properties can be derived. This approach retains full access to the electronic structure via the wavefunction at force field-like efficiency and captures quantum mechanics in an analytically differentiable representation. On several examples, we demonstrate that this opens promising avenues to perform inverse design of molecular structures for target electronic property optimisation and a clear path towards increased synergy of machine learning and quantum chemistry.

Deep learning has proven to yield fast and accurate predictions of quantum-chemical properties to accelerate the discovery of novel molecules and materials. As an exhaustive exploration of the vast chemical space is still infeasible, we require generative models that guide our search towards systems with desired properties. While graph-based models have previously been proposed, they are restricted by a lack of spatial information such that they are unable to recognize spatial isomerism and non-bonded interactions. Here, we introduce a generative neural network for 3d point sets that respects the rotational invariance of the targeted structures. We apply it to the generation of molecules and demonstrate its ability to approximate the distribution of equilibrium structures using spatial metrics as well as established measures from chemoinformatics. As our model is able to capture the complex relationship between 3d geometry and electronic properties, we bias the distribution of the generator towards molecules with a small HOMO-LUMO gap - an important property for the design of organic solar cells.

Viscous phenomena are the hallmark of the hydrodynamic flow exhibited by Dirac fermions in clean graphene at high enough temperatures. We report a quantitative calculation of the electronic shear and Hall viscosities in graphene based on the kinetic theory combined with the renormalization group providing a unified description at arbitrary doping levels and non-quantizing magnetic fields. At charge neutrality, the Hall viscosity vanishes, while the field-dependent shear viscosity decays from its zero-field value saturating to a nonzero value in classically strong fields. Away from charge neutrality, the field-dependent viscosity coefficients tend to agree with the semiclassical expectation., Comment: 14 pages, 8 figures, published version (minor typos corrected, references updated)

General solutions of state machine replication have to ensure that all replicas apply the same commands in the same order, even in the presence of failures. Such strict ordering incurs high synchronization costs caused by distributed consensus or by the use of a leader. This paper presents a protocol for linearizable state machine replication of conflict-free replicated data types (CRDTs) that neither requires consensus nor a leader. By leveraging the properties of state-based CRDTs - in particular, the monotonic growth of a join semilattice - synchronization overhead is greatly reduced. As a result, updates only need a single round trip and modify the state 'in-place' without the need for a log. Furthermore, the message size overhead for coordination consists of a single counter per message. For queries, we guarantee finite writes termination. We show in an experimental evaluation that more than 99 % of queries can be handled in one to three round trips under highly concurrent accesses. Our protocol achieves high throughput without auxiliary processes such as command log management or leader election. Thus, it is well suited for practical scenarios that need linearizable access to CRDT data on a fine-granular scale.

We introduce the notion of Loewner (ellipsoid) function for a log concave function and show that it is an extension of the Loewner ellipsoid for convex bodies. We investigate its duality relation to the recently defined John (ellipsoid) function by Alonso-Gutierrez, Merino, Jimenez and Villa. For convex bodies, John and Loewner ellipsoids are dual to each other. Interestingly, this need not be the case for the John function and the Loewner function.

We present timing solutions for four pulsars discovered in the Green Bank Northern Celestial Cap (GBNCC) survey. All four pulsars are isolated with spin periods between 0.26$\,$s and 1.84$\,$s. PSR J0038$-$2501 has a 0.26$\,$s period and a period derivative of ${7.6} \times {10}^{-19}\,{\rm s\,s}^{-1}$, which is unusually low for isolated pulsars with similar periods. This low period derivative may be simply an extreme value for an isolated pulsar or it could indicate an unusual evolution path for PSR J0038$-$2501, such as a disrupted recycled pulsar (DRP) from a binary system or an orphaned central compact object (CCO). Correcting the observed spin-down rate for the Shklovskii effect suggests that this pulsar may have an unusually low space velocity, which is consistent with expectations for DRPs. There is no X-ray emission detected from PSR J0038$-$2501 in an archival swift observation, which suggests that it is not a young orphaned CCO. The high dispersion measure of PSR J1949+3426 suggests a distance of 12.3$\,$kpc. This distance indicates that PSR J1949+3426 is among the most distant 7% of Galactic field pulsars, and is one of the most luminous pulsars., Comment: 7 pages, 5 figures

Dipolar interactions are crucial in the modeling of many complex magnetic systems, such as the pyrochlores and artificial spin systems. Remarkably, many classical dipolar coupled spin systems exhibit a continuous ground-state degeneracy, which is unexpected as the Hamiltonian does not possess a continuous symmetry. In this paper, we explain, how such a finite point-symmetry leads to a continuous ground-state degeneracy of specific classical dipolar-coupled systems. This work, therefore, provides new insight into the theory of classical dipolar-coupled spin-systems and opens the way to understand more complex dipolar coupled systems.

We study K3 surfaces with 9 cusps, i.e. 9 disjoint $A_2$ configurations of smooth rational curves, over algebraically closed fields of characteristic $p\neq 3$. Much like in the complex situation studied by Barth, we prove that each such surface admits a triple covering by an abelian surface. Conversely, we determine which abelian surfaces with order three automorphisms give rise to K3 surfaces. We also investigate how K3 surfaces with 9 cusps hit the supersingular locus., Comment: 17 pages

Deep Learning has been shown to learn efficient representations for structured data such as image, text or audio. In this chapter, we present neural network architectures that are able to learn efficient representations of molecules and materials. In particular, the continuous-filter convolutional network SchNet accurately predicts chemical properties across compositional and configurational space on a variety of datasets. Beyond that, we analyze the obtained representations to find evidence that their spatial and chemical properties agree with chemical intuition.

Discovery of atomistic systems with desirable properties is a major challenge in chemistry and material science. Here we introduce a novel, autoregressive, convolutional deep neural network architecture that generates molecular equilibrium structures by sequentially placing atoms in three-dimensional space. The model estimates the joint probability over molecular configurations with tractable conditional probabilities which only depend on distances between atoms and their nuclear charges. It combines concepts from state-of-the-art atomistic neural networks with auto-regressive generative models for images and speech. We demonstrate that the architecture is capable of generating molecules close to equilibrium for constitutional isomers of C$_7$O$_2$H$_{10}$.

In this work, we extend the SchNet architecture by using weighted skip connections to assemble the final representation. This enables us to study the relative importance of each interaction block for property prediction. We demonstrate on both the QM9 and MD17 dataset that their relative weighting depends strongly on the chemical composition and configurational degrees of freedom of the molecules which opens the path towards a more detailed understanding of machine learning models for molecules., Comment: NIPS 2018 Workshop: Machine Learning for Molecules and Materials

In this paper, we provide an accessible introduction to the theory of locally convex supermanifolds in the categorical approach. In this setting, a supermanifold is a functor $\mathcal{M}\colon\mathbf{Gr}\to\mathbf{Man}$ from the category of Grassmann algebras to the category of locally convex manifolds that has certain local models, forming something akin to an atlas. We give a mostly self-contained, concrete definition of supermanifolds along these lines, closing several gaps in the literature on the way. If $\Lambda_n\in\mathbf{Gr}$ is the Grassmann algebra with $n$ generators, we show that $\mathcal{M}_{\Lambda_n}$ has the structure of a so called multilinear bundle over the base manifold $\mathcal{M}_\mathbb{R}$. We use this fact to show that the projective limit $\varprojlim_n\mathcal{M}_{\Lambda_n}$ exists in the category of manifolds. In fact, this gives us a faithful functor $\varprojlim\colon\mathbf{SMan}\to\mathbf{Man}$ from the category of supermanifolds to the category of manifolds. This functor respects products, commutes with the respective tangent functor and retains the respective Hausdorff property. In this way, supermanifolds can be seen as a particular kind of infinite-dimensional fiber bundles., Comment: 58 pages, various minor corrections and changes

We exhibit three double octic Calabi--Yau threefolds over the certain quadratic fields and prove their modularity. The non-rigid threefold has two conjugate Hilbert modular forms of weight [4,2] and [2,4] attached while the two rigid threefolds correspond to a Hilbert modular form of weight [4,4] and to the twist of the restriction of a classical modular form of weight 4., Comment: 12 pages

Little known relations of the renown concept of the halfspace depth for multivariate data with notions from convex and affine geometry are discussed. Halfspace depth may be regarded as a measure of symmetry for random vectors. As such, the depth stands as a generalization of a measure of symmetry for convex sets, well studied in geometry. Under a mild assumption, the upper level sets of the halfspace depth coincide with the convex floating bodies used in the definition of the affine surface area for convex bodies in Euclidean spaces. These connections enable us to partially resolve some persistent open problems regarding theoretical properties of the depth.

SchNetPack is a toolbox for the development and application of deep neural networks to the prediction of potential energy surfaces and other quantum-chemical properties of molecules and materials. It contains basic building blocks of atomistic neural networks, manages their training and provides simple access to common benchmark datasets. This allows for an easy implementation and evaluation of new models. For now, SchNetPack includes implementations of (weighted) atomcentered symmetry functions and the deep tensor neural network SchNet as well as ready-to-use scripts that allow to train these models on molecule and material datasets. Based upon the PyTorch deep learning framework, SchNetPack allows to efficiently apply the neural networks to large datasets with millions of reference calculations as well as parallelize the model across multiple GPUs. Finally, SchNetPack provides an interface to the Atomic Simulation Environment in order to make trained models easily accessible to researchers that are not yet familiar with neural networks.

We compare the robustness of humans and current convolutional deep neural networks (DNNs) on object recognition under twelve different types of image degradations. First, using three well known DNNs (ResNet-152, VGG-19, GoogLeNet) we find the human visual system to be more robust to nearly all of the tested image manipulations, and we observe progressively diverging classification error-patterns between humans and DNNs when the signal gets weaker. Secondly, we show that DNNs trained directly on distorted images consistently surpass human performance on the exact distortion types they were trained on, yet they display extremely poor generalisation abilities when tested on other distortion types. For example, training on salt-and-pepper noise does not imply robustness on uniform white noise and vice versa. Thus, changes in the noise distribution between training and testing constitutes a crucial challenge to deep learning vision systems that can be systematically addressed in a lifelong machine learning approach. Our new dataset consisting of 83K carefully measured human psychophysical trials provide a useful reference for lifelong robustness against image degradations set by the human visual system., Comment: Added optimal probability aggregation method to appendix

In recent years, deep neural networks have revolutionized many application domains of machine learning and are key components of many critical decision or predictive processes. Therefore, it is crucial that domain specialists can understand and analyze actions and pre- dictions, even of the most complex neural network architectures. Despite these arguments neural networks are often treated as black boxes. In the attempt to alleviate this short- coming many analysis methods were proposed, yet the lack of reference implementations often makes a systematic comparison between the methods a major effort. The presented library iNNvestigate addresses this by providing a common interface and out-of-the- box implementation for many analysis methods, including the reference implementation for PatternNet and PatternAttribution as well as for LRP-methods. To demonstrate the versatility of iNNvestigate, we provide an analysis of image classifications for variety of state-of-the-art neural network architectures.

With the rise of deep neural networks for quantum chemistry applications, there is a pressing need for architectures that, beyond delivering accurate predictions of chemical properties, are readily interpretable by researchers. Here, we describe interpretation techniques for atomistic neural networks on the example of Behler-Parrinello networks as well as the end-to-end model SchNet. Both models obtain predictions of chemical properties by aggregating atom-wise contributions. These latent variables can serve as local explanations of a prediction and are obtained during training without additional cost. Due to their correspondence to well-known chemical concepts such as atomic energies and partial charges, these atom-wise explanations enable insights not only about the model but more importantly about the underlying quantum-chemical regularities. We generalize from atomistic explanations to 3d space, thus obtaining spatially resolved visualizations which further improve interpretability. Finally, we analyze learned embeddings of chemical elements that exhibit a partial ordering that resembles the order of the periodic table. As the examined neural networks show excellent agreement with chemical knowledge, the presented techniques open up new venues for data-driven research in chemistry, physics and materials science.

Nematicity is ubiquitous in electronic phases of high-$T_c$ superconductors, particularly in the Fe-based systems. While several experiments have probed nematic fluctuations, they have been restricted to uniform or momentum averaged fluctuations. Here, we investigate the behavior of finite-momentum nematic fluctuations by utilizing the anomalous softening of acoustic phonon modes in optimally doped Ba(Fe$_{0.94}$Co$_{0.06}$)$_2$As$_2$. We determine the nematic correlation length and find that it sharply changes its $T$-dependence at $T_c$, revealing a strong connection between nematicity and superconductivity.

We investigate weighted floating bodies of polytopes. We show that the weighted volume depends on the complete flags of the polytope. This connection is obtained by introducing flag simplices, which translate between the metric and combinatorial structure. Our results are applied in spherical and hyperbolic space. This leads to new asymptotic results for polytopes in these spaces. We also provide explicit examples of spherical and hyperbolic convex bodies whose floating bodies behave completely different from any convex body in Euclidean space.

Bottom-up and top-down, as well as low-level and high-level factors influence where we fixate when viewing natural scenes. However, the importance of each of these factors and how they interact remains a matter of debate. Here, we disentangle these factors by analysing their influence over time. For this purpose we develop a saliency model which is based on the internal representation of a recent early spatial vision model to measure the low-level bottom-up factor. To measure the influence of high-level bottom-up features, we use a recent DNN-based saliency model. To account for top-down influences, we evaluate the models on two large datasets with different tasks: first, a memorisation task and, second, a search task. Our results lend support to a separation of visual scene exploration into three phases: The first saccade, an initial guided exploration characterised by a gradual broadening of the fixation density, and an steady state which is reached after roughly 10 fixations. Saccade target selection during the initial exploration and in the steady state are related to similar areas of interest, which are better predicted when including high-level features. In the search dataset, fixation locations are determined predominantly by top-down processes. In contrast, the first fixation follows a different fixation density and contains a strong central fixation bias. Nonetheless, first fixations are guided strongly by image properties and as early as 200 ms after image onset, fixations are better predicted by high-level information. We conclude that any low-level bottom-up factors are mainly limited to the generation of the first saccade. All saccades are better explained when high-level features are considered, and later this high-level bottom-up control can be overruled by top-down influences., Comment: Submitted to Journal of Vision

We introduce certain rational functions on a smooth projective surface X in IP^3 which facilitate counting the lines on X. We apply this to smooth quintics in characteristic zero to prove that they contain no more than 127 lines, and that any given line meets at most 28 others. We construct examples which demonstrate that the latter bound is sharp., Comment: 30 pages; v2: minor edits following referee's comments

Multi-band superconductivity is realized in a plethora of systems, from high-temperature superconductors to very diluted superconductors. While several properties of multi-band superconductors can be understood as straightforward generalizations of their single-band counterparts, recent works have unveiled rather unusual behaviors unique to the former case. In this regard, a regime that has received significant attention is that near a Lifshitz transition, in which one of the bands crosses the Fermi level. In this work, we investigate how impurity scattering $\tau^{-1}$ affects the superconducting transition temperature $T_{c}$ across a Lifshitz transition, in the regime where intra-band pairing is dominant and inter-band pairing is subleading. This is accomplished by deriving analytic asymptotic expressions for $T_{c}$ and $\partial T_{c}/\partial\tau^{-1}$ in a two-dimensional two-band system. When the inter-band pairing interaction is repulsive, we find that, despite the incipient nature of the band crossing the Fermi level, inter-band impurity scattering is extremely effective in breaking Cooper pairs, making $\partial T_{c}/\partial\tau^{-1}$ quickly approach the limiting Abrikosov-Gor'kov value of the high-density regime. In contrast, when the inter-band pairing interaction is attractive, pair-breaking is much less efficient, affecting $T_{c}$ only mildly at the vicinity of the Lifshitz transition. The consequence of this general result is that the behavior of $T_{c}$ across a Lifshitz transition can be qualitatively changed in the presence of strong enough disorder: instead of displaying a sharp increase across the Lifshitz transition, as in the clean case, $T_{c}$ can actually display a maximum and be suppressed at the Lifshitz transition. These results shed new light on the non-trivial role of impurity scattering in multi-band superconductors., Comment: 17 pages, 12 figures

Although discovered many decades ago, superconductivity in doped SrTiO$_{3}$ remains a topic of intense research. Recent experiments revealed that, upon increasing the carrier concentration, multiple bands cross the Fermi level, signaling the onset of Lifshitz transitions. Interestingly, $T_{c}$ was observed to be suppressed across the Lifshitz transition of oxygen-deficient SrTiO$_{3}$; a similar behavior was also observed in gated LaAlO$_{3}$/SrTiO$_{3}$ interfaces. Such a behavior is difficult to explain in the clean theory of two-band superconductivity, as the additional electronic states provided by the second band should enhance $T_{c}$. Here, we show that this unexpected behavior can be explained by the strong pair-breaking effect promoted by disorder, which takes place if the inter-band pairing interaction is subleading and repulsive. A consequence of this scenario is that, upon moving away from the Lifshitz transition, the two-band superconducting state changes from opposite-sign gaps to same-sign gaps., Comment: 5 pages, 4 figures

When searching a target in a natural scene, both the target's visual properties and similarity to the background influence whether (and how fast) humans are able to find it. However, thus far it has been unclear whether searchers adjust the dynamics of their eye movements (e.g., fixation durations, saccade amplitudes) to the target they search for. In our experiment participants searched natural scenes for six artificial targets with different spatial frequency throughout eight consecutive sessions. High-spatial frequency targets led to smaller saccade amplitudes and shorter fixation durations than low-spatial frequency targets if target identity was known before the trial. If a saccade was programmed in the same direction as the previous saccade (saccadic momentum), fixation durations and successive saccade amplitudes were not influenced by target type. Visual saliency and empirical density at the endpoints of saccadic momentum saccades were comparatively low, indicating that these saccades were less selective. Our results demonstrate that searchers adjust their eye movement dynamics to the search target in a sensible fashion, since low-spatial frequencies are visible farther into the periphery than high-spatial frequencies. Additionally, the saccade direction specificity of our effects suggests a separation of saccades into a default scanning mechanism and a selective, target-dependent mechanism.

Deep learning has led to a paradigm shift in artificial intelligence, including web, text and image search, speech recognition, as well as bioinformatics, with growing impact in chemical physics. Machine learning in general and deep learning in particular is ideally suited for representing quantum-mechanical interactions, enabling to model nonlinear potential-energy surfaces or enhancing the exploration of chemical compound space. Here we present the deep learning architecture SchNet that is specifically designed to model atomistic systems by making use of continuous-filter convolutional layers. We demonstrate the capabilities of SchNet by accurately predicting a range of properties across chemical space for \emph{molecules and materials} where our model learns chemically plausible embeddings of atom types across the periodic table. Finally, we employ SchNet to predict potential-energy surfaces and energy-conserving force fields for molecular dynamics simulations of small molecules and perform an exemplary study of the quantum-mechanical properties of C$_{20}$-fullerene that would have been infeasible with regular ab initio molecular dynamics.

We introduce floating bodies for convex, not necessarily bounded subsets of $\mathbb{R}^n$. This allows us to define floating functions for convex and log concave functions and log concave measures. We establish the asymptotic behavior of the integral difference of a log concave function and its floating function. This gives rise to a new affine invariant which bears striking similarities to the Euclidean affine surface area.

Saliency methods aim to explain the predictions of deep neural networks. These methods lack reliability when the explanation is sensitive to factors that do not contribute to the model prediction. We use a simple and common pre-processing step ---adding a constant shift to the input data--- to show that a transformation with no effect on the model can cause numerous methods to incorrectly attribute. In order to guarantee reliability, we posit that methods should fulfill input invariance, the requirement that a saliency method mirror the sensitivity of the model with respect to transformations of the input. We show, through several examples, that saliency methods that do not satisfy input invariance result in misleading attribution.

We construct Zariski K3 surfaces of Artin invariant 1, 2 and 3 in many characteristics. In particular, we prove that any supersingular Kummer surface is Zariski if the characteristic is not congruent to 1 modulo 12. Our methods combine different approaches such as quotients by the group scheme $\alpha_p$, Kummer surfaces, and automorphisms of elliptic and hyperelliptic curves., Comment: 30 pages