Search

Text readability measures are widely used in many real-world scenarios and in NLP. These measures have primarily been developed by predicting reading comprehension outcomes, while largely neglecting what is perhaps the core aspect of a readable text: reading ease. In this work, we propose a new eye tracking based methodology for evaluating readability measures, which focuses on their ability to account for reading facilitation effects in text simplification, as well as for text reading ease more broadly. Using this approach, we find that existing readability formulas are moderate to poor predictors of reading ease. We further find that average per-word length, frequency, and especially surprisal tend to outperform existing readability formulas as measures of reading ease. We thus propose surprisal as a simple unsupervised alternative to existing measures.

Recent advancements in large vision-language models have enabled highly expressive and diverse vector sketch generation. However, state-of-the-art methods rely on a time-consuming optimization process involving repeated feedback from a pretrained model to determine stroke placement. Consequently, despite producing impressive sketches, these methods are limited in practical applications. In this work, we introduce SwiftSketch, a diffusion model for image-conditioned vector sketch generation that can produce high-quality sketches in less than a second. SwiftSketch operates by progressively denoising stroke control points sampled from a Gaussian distribution. Its transformer-decoder architecture is designed to effectively handle the discrete nature of vector representation and capture the inherent global dependencies between strokes. To train SwiftSketch, we construct a synthetic dataset of image-sketch pairs, addressing the limitations of existing sketch datasets, which are often created by non-artists and lack professional quality. For generating these synthetic sketches, we introduce ControlSketch, a method that enhances SDS-based techniques by incorporating precise spatial control through a depth-aware ControlNet. We demonstrate that SwiftSketch generalizes across diverse concepts, efficiently producing sketches that combine high fidelity with a natural and visually appealing style., Comment: https://swiftsketch.github.io/

The demand for low-power inference and training of deep neural networks (DNNs) on edge devices has intensified the need for algorithms that are both scalable and energy-efficient. While spiking neural networks (SNNs) allow for efficient inference by processing complex spatio-temporal dynamics in an event-driven fashion, training them on resource-constrained devices remains challenging due to the high computational and memory demands of conventional error backpropagation (BP)-based approaches. In this work, we draw inspiration from biological mechanisms such as eligibility traces, spike-timing-dependent plasticity, and neural activity synchronization to introduce TESS, a temporally and spatially local learning rule for training SNNs. Our approach addresses both temporal and spatial credit assignments by relying solely on locally available signals within each neuron, thereby allowing computational and memory overheads to scale linearly with the number of neurons, independently of the number of time steps. Despite relying on local mechanisms, we demonstrate performance comparable to the backpropagation through time (BPTT) algorithm, within $\sim1.4$ accuracy points on challenging computer vision scenarios relevant at the edge, such as the IBM DVS Gesture dataset, CIFAR10-DVS, and temporal versions of CIFAR10, and CIFAR100. Being able to produce comparable performance to BPTT while keeping low time and memory complexity, TESS enables efficient and scalable on-device learning at the edge., Comment: 9 pages, 2 figures

Compute-in-memory (CIM) accelerators for spiking neural networks (SNNs) are promising solutions to enable $\mu$s-level inference latency and ultra-low energy in edge vision applications. Yet, their current lack of flexibility at both the circuit and system levels prevents their deployment in a wide range of real-life scenarios. In this work, we propose a novel digital CIM macro that supports arbitrary operand resolution and shape, with a unified CIM storage for weights and membrane potentials. These circuit-level techniques enable a hybrid weight- and output-stationary dataflow at the system level to maximize operand reuse, thereby minimizing costly on- and off-chip data movements during the SNN execution. Measurement results of a fabricated FlexSpIM prototype in 40-nm CMOS demonstrate a 2$\times$ increase in bit-normalized energy efficiency compared to prior fixed-precision digital CIM-SNNs, while providing resolution reconfiguration with bitwise granularity. Our approach can save up to 90% energy in large-scale systems, while reaching a state-of-the-art classification accuracy of 95.8% on the IBM DVS gesture dataset., Comment: 5 pages, 7 figures, submitted to IEEE ISCAS 2025

The Voronoi Entropy (VE) and the continuous measure of symmetry (CSM) characterize the orderliness of a set of points on a 2D plane. The Voronoi entropy is the Shannon entropy of the Voronoi tessellation of the plane into polygons, quantifying the diversity of polygons. The VE is widely used to study the self-assembly of colloidal, supramolecular, and other systems. The value of VE changes from S=0 for a completely ordered system, built of polygons with an equal number of sides, to S=1.690+/-0.001 for a random set of points. While the VE takes into account only neighboring polygons, covering the 2D plane imposes constraints on the number of polygons and the number of edges in polygons. Consequently, unlike the conventional Shannon Entropy, the VE captures some long-range order properties of the system. We calculate the VE for several hyperuniform sets of points and compare it with the values of exponents of collective density variables characterizing long-range correlations in the system. We show that the VE correlates with the latter up to a certain saturation level, after which the value of the VE falls to S=0, and we explain this phenomenon.

We propose an off-shell bosonic string action that removes the renormalization window constraint of [1]. To all orders in conformal perturbation theory, this action allows for deformations of the worldsheet theory by any primary or descendant irrelevant deformation. Non-perturbatively, this action has no spurious solutions on the space of all worldsheet theories with a unitary matter sector that flows from a UV fixed point. We find that non-minimal couplings dressed with more than one factor of the Ricci curvature behave as gauge redundancies. As part of our investigation of this action, we find non-smooth behavior in the Zamolodchikov $C$-function. Our results mostly apply to Euclidean-signature target spaces, but can be extended to Lorentzian backgrounds which are invariant under time translations and CTO symmetry., Comment: 27 pages, 5 figures

We examine the Einstein-Cartan (EC) theory in first-order form, which has a diffeomorphism as well as a local Lorentz invariance. We study the renormalizability of this theory in the framework of the Batalin-Vilkovisky formalism, which allows for a gauge invariant renormalization. Using the background field method, we discuss the gauge invariance of the background effective action and analyze the Ward identities which reflect the symmetries of the EC theory. As an application, we compute, in a general background gauge, the self-energy of the tetrad field at one-loop order., Comment: 23 pages, 6 figures. To appear in Annals of Physics

First-order logic has been established as an important tool for modeling and verifying intricate systems such as distributed protocols and concurrent systems. These systems are parametric in the number of nodes in the network or the number of threads, which is finite in any system instance, but unbounded. One disadvantage of first-order logic is that it cannot distinguish between finite and infinite structures, leading to spurious counterexamples. To mitigate this, we offer a verification approach that captures only finite system instances. Our approach is an adaptation of the cutoff method to systems modeled in first-order logic. The idea is to show that any safety violation in a system instance of size larger than some bound can be simulated by a safety violation in a system of a smaller size. The simulation provides an inductive argument for correctness in finite instances, reducing the problem to showing safety of instances with bounded size. To this end, we develop a framework to (i) encode such simulation relations in first-order logic and to (ii) validate the simulation relation by a set of verification conditions given to an SMT solver. We apply our approach to verify safety of a set of examples, some of which cannot be proven by a first-order inductive invariant., Comment: 28 pages, 6 figures, 1 table

We define a relational notion of a subsystem in theories of matrix quantum mechanics and show how the corresponding entanglement entropy can be given as a minimisation, exhibiting many similarities to the Ryu-Takayanagi formula. Our construction brings together the physics of entanglement edge modes, noncommutative geometry and quantum internal reference frames, to define a subsystem whose reduced state is (approximately) an incoherent sum of density matrices, corresponding to distinct spatial subregions. We show that in states where geometry emerges from semiclassical matrices, this sum is dominated by the subregion with minimal boundary area. As in the Ryu-Takayanagi formula, it is the computation of the entanglement that determines the subregion. We find that coarse-graining is essential in our microscopic derivation, in order to control the proliferation of highly curved and disconnected non-geometric subregions in the sum., Comment: 48+16 pages, 8 figures

This paper introduces a novel computational approach for analyzing nonverbal social behavior in educational settings. Integrating multimodal behavioral cues, including facial expressions, gesture intensity, and spatial dynamics, the model assesses the nonverbal immediacy (NVI) of teachers from RGB classroom videos. A dataset of 400 30-second video segments from German classrooms was constructed for model training and validation. The gesture intensity regressor achieved a correlation of 0.84, the perceived distance regressor 0.55, and the NVI model 0.44 with median human ratings. The model demonstrates the potential to provide a valuable support in nonverbal behavior assessment, approximating the accuracy of individual human raters. Validated against both questionnaire data and trained observer ratings, our models show moderate to strong correlations with relevant educational outcomes, indicating their efficacy in reflecting effective teaching behaviors. This research advances the objective assessment of nonverbal communication behaviors, opening new pathways for educational research., Comment: 12 pages, 3 figures. Camera-ready version for the SAB 2024: 17th International Conference on the Simulation of Adaptive Behavior

We propose a simple connection between matrix quantum mechanics and tensor networks. This allows us to imbue tensor networks with some interesting additional structure. The geometry of the graph describing the tensor network state is determined dynamically, giving a notion of background independence. The tensor network states have a $U(N)$ invariance, which (a) allows us to consider continuous families of entanglement cuts even with a finite number of tensors and (b) includes a notion of bulk coordinate reparameterization and area-preserving diffeomorphism invariance in the large N limit. These tensor networks also have a natural scale of nonlocality that behaves similarly to a string scale, suggesting a potential toy model for sub-AdS physics. Emergent $U(p)$ gauge fields naturally appear on the tensor network links., Comment: 32 pages, 19 figures

Supported metal nanoparticle (NP) catalysts are vital for the sustainable production of chemicals, but their design and implementation are limited by the ability to identify and characterize their structures and atomic sites that are correlated with high catalytic activity. Identification of these ''active sites'' has relied heavily on extrapolation to supported NP systems from investigation of idealized surfaces, experimentally using single crystals or supported NPs which are always modelled computationally using slab or regular polyhedra models. However, the ability of these methods to predict catalytic activity remains qualitative at best, as the structure of metal NPs in reactive environments has only been speculated from indirect experimental observations, or otherwise remains wholly unknown. Here, by circumventing these limitations for highly accurate simulation methods, we provide direct atomistic insight into the dynamic restructuring of metal NPs by combining in situ spectroscopy with molecular dynamics simulations powered by a machine learned force field. We find that in reactive environments, NP surfaces evolve to a state with poorly defined atomic order, while the core of the NP may remain bulk-like. These insights prove that long-standing conceptual models based on idealized faceting for small metal NP systems are not representative of real systems under exposure to reactive environments. We show that the resultant structure can be elucidated by combining advanced spectroscopy and computational tools. This discovery exemplifies that to enable faithful quantitative predictions of catalyst function and stability, we must move beyond idealized-facet experimental and theoretical models and instead employ systems which include realistic surface structures that respond to relevant physical and chemical conditions., Comment: 18 pages, 11 figures (4 main, 7 SI)

We introduce the triangulant of two matrices, and relate it to the existence of orthogonal eigenvectors. We also use it for a new characterization of mutually unbiased bases. Generalizing the notion, we introduce higher order triangulants of two matrices, and relate them to the existence of nontrivially intersecting invariant subspaces of complementary dimensions., Comment: 16 pages

With the rise of artificial intelligence, neural network simulations of biological neuron models are being explored to reduce the footprint of learning and inference in resource-constrained task scenarios. A mainstream type of such networks are spiking neural networks (SNNs) based on simplified Integrate and Fire models for which several hardware accelerators have emerged. Among them, the ReckOn chip was introduced as a recurrent SNN allowing for both online training and execution of tasks based on arbitrary sensory modalities, demonstrated for vision, audition, and navigation. As a fully digital and open-source chip, we adapted ReckOn to be implemented on a Xilinx Multiprocessor System on Chip system (MPSoC), facilitating its deployment in embedded systems and increasing the setup flexibility. We present an overview of the system, and a Python framework to use it on a Pynq ZU platform. We validate the architecture and implementation in the new scenario of robotic arm control, and show how the simulated accuracy is preserved with a peak performance of 3.8M events processed per second., Comment: Under review at ICECS'24

We present an automata-based algorithm to synthesize omega-regular causes for omega-regular effects on executions of a reactive system, such as counterexamples uncovered by a model checker. Our theory is a generalization of temporal causality, which has recently been proposed as a framework for drawing causal relationships between trace properties on a given trace. So far, algorithms exist only for verifying a single causal relationship and, as an extension, cause synthesis through enumeration, which is complete only for a small fragment of effect properties. This work presents the first complete cause-synthesis algorithm for the class of omega-regular effects. We show that in this case, causes are guaranteed to be omega-regular themselves and can be computed as, e.g., nondeterministic B\"uchi automata. We demonstrate the practical feasibility of this algorithm with a prototype tool and evaluate its performance for cause synthesis and cause checking., Comment: Full version with appendix

This paper lays a practical foundation for using abstract interpretation with an abstract domain that consists of sets of quantified first-order logic formulas. This abstract domain seems infeasible at first sight due to the complexity of the formulas involved and the enormous size of sets of formulas (abstract elements). We introduce an efficient representation of abstract elements, which eliminates redundancies based on a novel syntactic subsumption relation that under-approximates semantic entailment. We develop algorithms and data structures to efficiently compute the join of an abstract element with the abstraction of a concrete state, operating on the representation of abstract elements. To demonstrate feasibility of the domain, we use our data structures and algorithms to implement a symbolic abstraction algorithm that computes the least fixpoint of the best abstract transformer of a transition system, which corresponds to the strongest inductive invariant. We succeed at finding, for example, the least fixpoint for Paxos (which in our representation has 1,438 formulas with $\forall^*\exists^*\forall^*$ quantification) in time comparable to state-of-the-art property-directed approaches.

Edge vision systems combining sensing and embedded processing promise low-latency, decentralized, and energy-efficient solutions that forgo reliance on the cloud. As opposed to conventional frame-based vision sensors, event-based cameras deliver a microsecond-scale temporal resolution with sparse information encoding, thereby outlining new opportunities for edge vision systems. However, mainstream algorithms for frame-based vision, which mostly rely on convolutional neural networks (CNNs), can hardly exploit the advantages of event-based vision as they are typically optimized for dense matrix-vector multiplications. While event-driven graph neural networks (GNNs) have recently emerged as a promising solution for sparse event-based vision, their irregular structure is a challenge that currently hinders the design of efficient hardware accelerators. In this paper, we propose EvGNN, the first event-driven GNN accelerator for low-footprint, ultra-low-latency, and high-accuracy edge vision with event-based cameras. It relies on three central ideas: (i) directed dynamic graphs exploiting single-hop nodes with edge-free storage, (ii) event queues for the efficient identification of local neighbors within a spatiotemporally decoupled search range, and (iii) a novel layer-parallel processing scheme enabling the low-latency execution of multi-layer GNNs. We deployed EvGNN on a Xilinx KV260 Ultrascale+ MPSoC platform and benchmarked it on the N-CARS dataset for car recognition, demonstrating a classification accuracy of 87.8% and an average latency per event of 16$\mu$s, thereby enabling real-time, microsecond-resolution event-based vision at the edge., Comment: 12 pages, 14 figures

While the human brain efficiently adapts to new tasks from a continuous stream of information, neural network models struggle to learn from sequential information without catastrophically forgetting previously learned tasks. This limitation presents a significant hurdle in deploying edge devices in real-world scenarios where information is presented in an inherently sequential manner. Active dendrites of pyramidal neurons play an important role in the brain ability to learn new tasks incrementally. By exploiting key properties of time-to-first-spike encoding and leveraging its high sparsity, we present a novel spiking neural network model enhanced with active dendrites. Our model can efficiently mitigate catastrophic forgetting in temporally-encoded SNNs, which we demonstrate with an end-of-training accuracy across tasks of 88.3% on the test set using the Split MNIST dataset. Furthermore, we provide a novel digital hardware architecture that paves the way for real-world deployment in edge devices. Using a Xilinx Zynq-7020 SoC FPGA, we demonstrate a 100-% match with our quantized software model, achieving an average inference time of 37.3 ms and an 80.0% accuracy., Comment: This work was accepted and presented at AICAS 2024

In lead halide perovskites (APbX3), the effect of the A-site cation on optical and electronic properties has initially been thought to be marginal. Yet, evidence of beneficial effects on solar cell performance and light emission is accumulating. Here, we report that the A-cation in soft APbBr3 colloidal quantum dots (QDs) controls the phonon-induced localization of the exciton wavefunction. Insights from ab initio molecular dynamics and single-particle fluorescence spectroscopy demonstrate that anharmonic lattice vibrations and the resulting polymorphism act as an additional confinement potential. Avoiding the trade-off between single-photon purity and optical stability faced by downsizing conventional QDs into the strong confinement regime, dynamical phonon-induced confinement in large organic-inorganic perovskite QDs enables bright (10^6 photons/s), stable (> 1h), and pure (> 95%) single-photon emission in a widely tuneable spectral range (495-745 nm). Strong electron-phonon interaction in soft perovskite QDs provides an unconventional route toward the development of scalable room-temperature quantum light sources.

Hyperproperties express the relationship between multiple executions of a system. This is needed in many AI-related fields, such as knowledge representation and planning, to capture system properties related to knowledge, information flow, and privacy. In this paper, we study the monitoring of complex hyperproperties at runtime. Previous work in this area has either focused on the simpler problem of monitoring trace properties (which are sets of traces, while hyperproperties are sets of sets of traces) or on monitoring first-order hyperproperties, which are expressible in temporal logics with first-order quantification over traces, such as HyperLTL. We present the first monitoring algorithm for the much more expressive class of second-order hyperproperties. Second-order hyperproperties include system properties like common knowledge, which cannot be expressed in first-order logics like HyperLTL. We introduce Hyper$^2$LTL$_f$, a temporal logic over finite traces that allows for second-order quantification over sets of traces. We study the monitoring problem in two fundamental execution models: (1) the parallel model, where a fixed number of traces is monitored in parallel, and (2) the sequential model, where an unbounded number of traces is observed sequentially, one trace after the other. For the parallel model, we show that the monitoring of the second-order hyperproperties of Hyper$^2$LTL$_f$ can be reduced to monitoring first-order hyperproperties. For the sequential model, we present a monitoring algorithm that handles second-order quantification efficiently, exploiting optimizations based on the monotonicity of subformulas, graph-based storing of executions, and fixpoint hashing. We present experimental results from a range of benchmarks, including examples from common knowledge and planning., Comment: AAMAS 2024

Image stylization involves manipulating the visual appearance and texture (style) of an image while preserving its underlying objects, structures, and concepts (content). The separation of style and content is essential for manipulating the image's style independently from its content, ensuring a harmonious and visually pleasing result. Achieving this separation requires a deep understanding of both the visual and semantic characteristics of images, often necessitating the training of specialized models or employing heavy optimization. In this paper, we introduce B-LoRA, a method that leverages LoRA (Low-Rank Adaptation) to implicitly separate the style and content components of a single image, facilitating various image stylization tasks. By analyzing the architecture of SDXL combined with LoRA, we find that jointly learning the LoRA weights of two specific blocks (referred to as B-LoRAs) achieves style-content separation that cannot be achieved by training each B-LoRA independently. Consolidating the training into only two blocks and separating style and content allows for significantly improving style manipulation and overcoming overfitting issues often associated with model fine-tuning. Once trained, the two B-LoRAs can be used as independent components to allow various image stylization tasks, including image style transfer, text-based image stylization, consistent style generation, and style-content mixing., Comment: Project Page: https://b-lora.github.io/B-LoRA/

We study a new class of functions that arise naturally in quaternionic analysis, we call them "quasi regular functions". Like the well-known quaternionic regular functions, these functions provide representations of the quaternionic conformal group. However, unlike the regular functions, the quasi regular ones do not admit an invariant unitary structure but rather a pseudounitary equivalent. The reproducing kernels of these functions have an especially simple form: (Z-W)^{-1}. We describe the K-type bases of quasi regular functions and derive the reproducing kernel expansions. We also show that the restrictions of the irreducible representations formed from the quasi regular functions to the Poincare group have three irreducible components. Our interest in the quasi regular functions arises from an application to the study of conformal-invariant algebras of quaternionic functions. We also introduce a factorization of certain intertwining operators between tensor products of spaces of quaternionic functions. This factorization is obtained using fermionic Fock spaces constructed from the quasi regular functions., Comment: 65 pages, submitted

We study periodic infinite billiards in the plane. We show that for rational models, some particular obstacles can be added periodically, so that the billiard flow in the resulting table is recurrent in almost every direction.

We consider the Einstein-Cartan theory with the tetrad $e_{\mu}^{a}$ and spin connection $\omega_{\mu ab}$ taken as being independent fields. Diffeomorphism invariance and local Lorentz invariance result in there being two distinct gauge transformations in this approach, and consequently two ghost fields arise when employing the usual Faddeev-Popov quantization procedure. Our choice of gauge fixing retains the gauge invariances of the background field. We show that the gauge algebra is closed even in the presence of torsion, and the resulting BRST invariance can be found for the effective action. We also derive the Slavnov-Taylor identities, which reflect the BRST symmetries of this theory., Comment: 18 pages. To be published in Annals of Physics

Generalized Baxter's TQ-relations and the QQ-system are systems of algebraic relations in the category O of representations of the Borel subalgebra of the quantum affine algebra U_q(g^), which we established in our earlier works arXiv:1308.3444 and arXiv:1606.05301. In the present paper, we conjecture a family of analogous relations labeled by elements of the Weyl group W of g, so that the original relations correspond to the identity element. These relations are closely connected to the W-symmetry of q-characters established in arXiv:2211.09779. We prove these relations for all w in W if g has rank two, and we prove the extended TQ-relations if w is a simple reflection. We also generalize our results and conjectures to the shifted quantum affine algebras., Comment: 42 pages, v3: more details added, accepted for publication in Communications in Math. Phys

Graph decompositions are the natural generalisation of tree decompositions where the decomposition tree is replaced by a genuine graph. Recently they found theoretical applications in the theory of sparsity, topological graph theory, structural graph theory and geometric group theory. We demonstrate applicability of graph decompositions on large networks by implementing an efficient algorithm and testing it on road networks.