Search

Efficient long-context language modeling remains a significant challenge in Natural Language Processing (NLP). While Transformers dominate language tasks, they struggle with long sequences due to quadratic computational complexity in training and linearly scaling memory costs during inference. Recent State Space Models (SSMs) such as Mamba offer alternatives with constant memory usage, but they underperform in tasks requiring extensive in-context retrieval. We introduce Taipan, a novel hybrid architecture that combines Mamba-2 with Selective Attention Layers (SALs). These SALs identify tokens requiring long-range interactions, remove less important features, and then augment their representations using the attention module. This approach balances Mamba's efficiency with Transformer-like performance in memory-intensive tasks. By constraining the attention budget, Taipan extends accurate predictions to context lengths of up to 1 million tokens while preserving computational efficiency. Our experiments demonstrate Taipan's superior performance across various scales and tasks, offering a promising solution for efficient long-context language modeling.

The performance of superconducting qubits is often limited by dissipation and two-level systems (TLS) losses. The dominant sources of these losses are believed to originate from amorphous materials and defects at interfaces and surfaces, likely as a result of fabrication processes or ambient exposure. Here, we explore a novel wet chemical surface treatment at the Josephson junction-substrate and the substrate-air interfaces by replacing a buffered oxide etch (BOE) cleaning process with one that uses hydrofluoric acid followed by aqueous ammonium fluoride. We show that the ammonium fluoride etch process results in a statistically significant improvement in median $\text{T}_1$ by $\sim22\%$ ($p=0.002$), and a reduction in the number of strongly-coupled TLS in the tunable frequency range. Microwave resonator measurements on samples treated with the ammonium fluoride etch prior to niobium deposition also show $\sim33\%$ lower TLS-induced loss tangent compared to the BOE treated samples. As the chemical treatment primarily modifies the Josephson junction-substrate interface and substrate-air interface, we perform targeted chemical and structural characterizations to examine materials' differences at these interfaces and identify multiple microscopic changes that could contribute to decreased TLS.

In this paper, we study the distance problem in the setting of finite p-adic rings. In odd dimensions, our results are essentially sharp. In even dimensions, we clarify the conjecture and provide examples to support it. Surprisingly, compared to the finite field case, in this setting, we are able to provide a large family of sets such that the distance conjecture holds. By developing new restriction type estimates associated to circles and orbits, with a group theoretic argument, we will prove the $4/3$-parallel result in the two dimensions. This answers a question raised by Alex Iosevich. In a more general scenario, the existence/distribution of geometric/graph configurations will be also considered in this paper. Our results present improvements and extensions of recent results due to Ben Lichtin (2019, 2023). In comparison with Lichtin's method, our approach is much simpler and flexible, which is also one of the novelties in this paper., Comment: V3: typos corrected, 32 pages, final version

For Borel subsets $\Theta\subset O(d)\times \mathbb{R}^d$ (the set of all rigid motions) and $E\subset \mathbb{R}^d$, we define \begin{align*} \Theta(E):=\bigcup_{(g,z)\in \Theta}(gE+z). \end{align*} In this paper, we investigate the Lebesgue measure and Hausdorff dimension of $\Theta(E)$ given the dimensions of the Borel sets $E$ and $\Theta$, when $\Theta$ has product form. We also study this question by replacing rigid motions with the class of dilations and translations; and similarity transformations. The dimensional thresholds are sharp. Our results are variants of some previously known results in the literature when $E$ is restricted to smooth objects such as spheres, $k$-planes, and surfaces., Comment: 27 pages

In this paper, we introduce a new and direct approach to study the solvability of systems of equations generated by bilinear forms. More precisely, let $B (\cdot, \cdot)$ be a non-degenerate bilinear form and $E$ be a set in $\mathbb{F}_q^2$. We prove that if $|E|\gg q^{5/3}$ then the number of triples $(B(x, y), B(y, z), B(z, x))$ with $x, y, z\in E$ is at least $cq^3$ for some positive constant $c$. This significantly improves a result due to the fifth listed author (2009)., Comment: 12 pages

Superconducting Nb thin films have recently attracted significant attention due to their utility for quantum information technologies. In the processing of Nb thin films, fluoride-based chemical etchants are commonly used to remove surface oxides that are known to affect superconducting quantum devices adversely. However, these same etchants can also introduce hydrogen to form Nb hydrides, potentially negatively impacting microwave loss performance. Here, we present comprehensive materials characterization of Nb hydrides formed in Nb thin films as a function of fluoride chemical treatments. In particular, secondary-ion mass spectrometry, X-ray scattering, and transmission electron microscopy reveal the spatial distribution and phase transformation of Nb hydrides. The rate of hydride formation is determined by the fluoride solution acidity and the etch rate of Nb2O5, which acts as a diffusion barrier for hydrogen into Nb. The resulting Nb hydrides are detrimental to Nb superconducting properties and lead to increased power-independent microwave loss in coplanar waveguide resonators. However, Nb hydrides do not correlate with two-level system loss or device aging mechanisms. Overall, this work provides insight into the formation of Nb hydrides and their role in microwave loss, thus guiding ongoing efforts to maximize coherence time in superconducting quantum devices.

CLIP-based classifiers rely on the prompt containing a {class name} that is known to the text encoder. Therefore, they perform poorly on new classes or the classes whose names rarely appear on the Internet (e.g., scientific names of birds). For fine-grained classification, we propose PEEB - an explainable and editable classifier to (1) express the class name into a set of text descriptors that describe the visual parts of that class; and (2) match the embeddings of the detected parts to their textual descriptors in each class to compute a logit score for classification. In a zero-shot setting where the class names are unknown, PEEB outperforms CLIP by a huge margin (~10x in top-1 accuracy). Compared to part-based classifiers, PEEB is not only the state-of-the-art (SOTA) on the supervised-learning setting (88.80% and 92.20% accuracy on CUB-200 and Dogs-120, respectively) but also the first to enable users to edit the text descriptors to form a new classifier without any re-training. Compared to concept bottleneck models, PEEB is also the SOTA in both zero-shot and supervised-learning settings., Comment: Findings of NAACL 2024 (long paper)

We give an upper bound on the number of exceptional orthogonal projections of a small set of points in a plane over a prime order field, which makes progress toward a conjecture of Chen (2018). This theorem relies on a new upper bound on the number of incidences between an arbitrary set of lines and a set of points contained in a reasonably small pencil of lines. We also prove a new upper bound on the number of incidences between points and lines when both are Cartesian products in an appropriate sense.

A conjecture on radial projections in $\mathbb{F}_q^d$ states that if $E\subset \mathbb{F}_q^n$ with $q^{k-1}< |E|\leq q^{k}$ for some $k \in \{1,\dots, n-1\}$, then \[ \#\{y\in \mathbb{F}_q^n \mid |\pi^y(E)| < |E| \} \leq q^k. \] The case $k=n-1$ was recently proved by the second and third listed authors and Thu (2022). In this paper, we resolve this conjecture in all ranges of $k$.

Precise 3D environmental mapping is pivotal in robotics. Existing methods often rely on predefined concepts during training or are time-intensive when generating semantic maps. This paper presents Open-Fusion, a groundbreaking approach for real-time open-vocabulary 3D mapping and queryable scene representation using RGB-D data. Open-Fusion harnesses the power of a pre-trained vision-language foundation model (VLFM) for open-set semantic comprehension and employs the Truncated Signed Distance Function (TSDF) for swift 3D scene reconstruction. By leveraging the VLFM, we extract region-based embeddings and their associated confidence maps. These are then integrated with 3D knowledge from TSDF using an enhanced Hungarian-based feature-matching mechanism. Notably, Open-Fusion delivers outstanding annotation-free 3D segmentation for open-vocabulary without necessitating additional 3D training. Benchmark tests on the ScanNet dataset against leading zero-shot methods highlight Open-Fusion's superiority. Furthermore, it seamlessly combines the strengths of region-based VLFM and TSDF, facilitating real-time 3D scene comprehension that includes object concepts and open-world semantics. We encourage the readers to view the demos on our project page: https://uark-aicv.github.io/OpenFusion

We have combined the benefits of two catalytic growth phenomena to form nanostructures of transition metal trichalcogenides (TMTs), materials that are challenging to grow in a nanostructured form by conventional techniques, as required to exploit their exotic physics. Our growth strategy combines the benefits of vapor-liquid-solid (VLS) growth in controlling dimension and growth location, and salt-assisted growth for fast growth at moderate temperatures. This salt-assisted VLS growth is enabled through use of a catalyst that includes Au and an alkali metal halide. We demonstrate high yields of NbS3 1D nanostructures with sub-ten nanometer diameter, tens of micrometers length, and distinct 1D morphologies consisting of nanowires and nanoribbons with [010] and [100] growth orientations, respectively. We present strategies to control the growth location, size, and morphology. We extend the growth method to synthesize other TMTs, NbSe3 and TiS3, as nanowires. Finally, we discuss the growth mechanism based on the relationships we measure between the materials characteristics (growth orientation, morphology and dimensions) and the growth conditions (catalyst volume and growth time). Our study introduces opportunities to expand the library of emerging 1D vdW materials and their heterostructures with controllable nanoscale dimensions., Comment: 16 pages, 5 figures

This paper aims to introduce a more general definition of quasirandom groups and generalize several well-known results in the literature in this new setting. More precisely, let $G$ be a semi-direct product of groups and $X\subseteq G$, we provide conditions such that one can find tuples $(x_0, \ldots, x_k)\in X^{k+1}$ satisfying $x_1x_2\ldots x_k=x_0$ or conditions to guarantee that the product set $XX$ grows exponentially. In a special case of the group of rigid-motions in the plane over an arbitrary finite field, our results offer a reasonably complete description of structures of this group., Comment: V2: references updated

The first goal of this paper is to prove a sharp condition to guarantee of having a positive proportion of all congruence classes of triangles in given sets in $\mathbb{F}_q^2$. More precisely, for $A, B, C\subset \mathbb{F}_q^2$, if $|A||B||C|^{1/2}\gg q^4$, then for any $\lambda\in \mathbb{F}_q\setminus \{0\}$, the number of congruence classes of triangles with vertices in $A\times B\times C$ and one side-length $\lambda$ is at least $\gg q^2$. In higher dimensions, we obtain similar results for $k$-simplex but under a slightly stronger condition. Compared to the well--known $L^2$ method in the literature, our approach offers better results in both conditions and conclusions. When $A=B=C$, the second goal of this paper is to give a new and unified proof of the best current results on the distribution of simplex due to Bennett, Hart, Iosevich, Pakianathan, and Rudnev (2017) and McDonald (2020). The third goal of this paper is to study a Furstenberg-type problem associated to a set of rigid motions. The main ingredients in our proofs are incidence bounds between points and rigid motions. While the incidence bounds for large sets are due to the author and Semin Yoo (2023), the bound for small sets will be proved by using a point--line incidence bound in $\mathbb{F}_q^3$ due to Koll\'{a}r (2015)., Comment: V3: minor changes, typos corrected

Let $P$ be a $\delta$-separated $(\delta, s, C_P)$-set of points in $B(0, 1)\subset \mathbb{R}^d$ and $\Pi$ be a $\delta$-separated $(\delta, t, C_\Pi)$-set of hyperplanes intersecting $B(0, 1)$ in $\mathbb{R}^d$. Define \[I_{C\delta}(P, \Pi)=\#\{(p, \pi)\in P\times \Pi\colon p\in \pi(C\delta)\}.\] Suppose that $s, t\ge \frac{d+1}{2}$, then we have $I_{C\delta}(P, \Pi)\lesssim \delta |P||\Pi|$. The main ingredient in our argument is a measure theoretic result due to Eswarathansan, Iosevich, and Taylor (2011) which was proved by using Sobolev bounds for generalized Radon transforms. Our result is essentially sharp, a construction will be provided and discussed in the last section., Comment: 14 pages

In this paper, we study the distribution of parallelograms and rhombi in a given set in the plane over arbitrary finite fields $\mathbb{F}_q^2$. As an application, we improve a recent result due to Fitzpatrick, Iosevich, McDonald, and Wyman (2021) on the Vapnik-Chervonenkis dimension of the induced distance graph. Our proofs are based on the discrete Fourier analysis., Comment: 9 pages

Let $A$ and $B$ be sets in a finite vector space. In this paper, we study the magnitude of the set $A\cap f(B)$, where $f$ runs through a set of transformations. More precisely, we will focus on the cases that the set of transformations is given by orthogonal matrices or orthogonal projections. One of the most important contributions of this paper is to show that if $A, B\subset \mathbb{F}_q^d$ satisfy some natural conditions then, for almost every $g\in O(d)$, there are at least $\gg q^d$ elements $z\in \mathbb{F}_q^d$ such that \[|A\cap (g(B)+z)| \sim \frac{|A||B|}{q^d}.\] This infers that $|A-gB|\gg q^d$ for almost every $g\in O(d)$. In the flavor of expanding functions, with $|A|\le |B|$, we also show that the image $A-gB$ grows exponentially. In two dimensions, the result simply says that if $|A|=q^x$ and $|B|=q^y$, as long as $00$ such that $|A-gB|\gg |B|^{1+\epsilon}$. To prove these results, we need to develop a new and robust incidence bound between points and rigid motions by using a number of techniques including algebraic methods and discrete Fourier analysis. Our results are essentially sharp in odd dimensions., Comment: Submitted version. 40 pages

The advent of deep learning has led to a significant gain in machine translation. However, most of the studies required a large parallel dataset which is scarce and expensive to construct and even unavailable for some languages. This paper presents a simple yet effective method to tackle this problem for low-resource languages by augmenting high-quality sentence pairs and training NMT models in a semi-supervised manner. Specifically, our approach combines the cross-entropy loss for supervised learning with KL Divergence for unsupervised fashion given pseudo and augmented target sentences derived from the model. We also introduce a SentenceBERT-based filter to enhance the quality of augmenting data by retaining semantically similar sentence pairs. Experimental results show that our approach significantly improves NMT baselines, especially on low-resource datasets with 0.46--2.03 BLEU scores. We also demonstrate that using unsupervised training for augmented data is more efficient than reusing the ground-truth target sentences for supervised learning., Comment: TMP and GN contributed equally

Music-driven choreography is a challenging problem with a wide variety of industrial applications. Recently, many methods have been proposed to synthesize dance motions from music for a single dancer. However, generating dance motion for a group remains an open problem. In this paper, we present $\rm AIOZ-GDANCE$, a new large-scale dataset for music-driven group dance generation. Unlike existing datasets that only support single dance, our new dataset contains group dance videos, hence supporting the study of group choreography. We propose a semi-autonomous labeling method with humans in the loop to obtain the 3D ground truth for our dataset. The proposed dataset consists of 16.7 hours of paired music and 3D motion from in-the-wild videos, covering 7 dance styles and 16 music genres. We show that naively applying single dance generation technique to creating group dance motion may lead to unsatisfactory results, such as inconsistent movements and collisions between dancers. Based on our new dataset, we propose a new method that takes an input music sequence and a set of 3D positions of dancers to efficiently produce multiple group-coherent choreographies. We propose new evaluation metrics for measuring group dance quality and perform intensive experiments to demonstrate the effectiveness of our method. Our project facilitates future research on group dance generation and is available at: https://aioz-ai.github.io/AIOZ-GDANCE/, Comment: accepted in CVPR 2023

Let $G$ be a graph and $U\subset V(G)$ be a set of vertices. For each $v\in U$, let $h_v\colon U\to \{0, 1\}$ be the function defined by \[h_v(u)=\begin{cases} &1 ~\mbox{if}~u\sim v, u\in U\\&0 ~\mbox{if}~u\not\sim v, u\in U\end{cases},\] and set $\mathcal{H}(U):=\{h_v\colon v\in U\}$. The first purpose of this paper is to study the following question: What families of graphs $G$ and what conditions on $U$ do we need so that the VC-dimension of $\mathcal{H}(U)$ can be determined? We show that if $G$ is a pseudo-random graph, then under some mild conditions, the VC dimension of $\mathcal{H}(U)$ can be bounded from below. Specific cases of this theorem recover and improve previous results on VC-dimension of functions defined by the well-studied distance and dot-product graphs over a finite field.

The main purpose of this paper is to prove that the point-line incidence bound due to Vinh (2011) over arbitrary finite fields can be improved in certain ranges by using tools from the VC-dimension theory. As consequences, a number of applications will be discussed in detail., Comment: 14 pages

Data-driven materials design often encounters challenges where systems require or possess qualitative (categorical) information. Metal-organic frameworks (MOFs) are an example of such material systems. The representation of MOFs through different building blocks makes it a challenge for designers to incorporate qualitative information into design optimization. Furthermore, the large number of potential building blocks leads to a combinatorial challenge, with millions of possible MOFs that could be explored through time consuming physics-based approaches. In this work, we integrated Latent Variable Gaussian Process (LVGP) and Multi-Objective Batch-Bayesian Optimization (MOBBO) to identify top-performing MOFs adaptively, autonomously, and efficiently without any human intervention. Our approach provides three main advantages: (i) no specific physical descriptors are required and only building blocks that construct the MOFs are used in global optimization through qualitative representations, (ii) the method is application and property independent, and (iii) the latent variable approach provides an interpretable model of qualitative building blocks with physical justification. To demonstrate the effectiveness of our method, we considered a design space with more than 47,000 MOF candidates. By searching only ~1% of the design space, LVGP-MOBBO was able to identify all MOFs on the Pareto front and more than 97% of the 50 top-performing designs for the CO$_2$ working capacity and CO$_2$/N$_2$ selectivity properties. Finally, we compared our approach with the Random Forest algorithm and demonstrated its efficiency, interpretability, and robustness., Comment: 35 pages total. First 29 pages belong to the main manuscript and the remaining 6 six are for the supplementary information, 13 figures total. 9 figures are on the main manuscript and 4 figures are in the supplementary information. 1 table in the supplementary information

The purpose of this paper is to introduce and study the following graph theoretic paradigm. Let $$T_Kf(x)=\int K(x,y) f(y) d\mu(y),$$ where $f: X \to {\Bbb R}$, $X$ a set, finite or infinite, and $K$ and $\mu$ denote a suitable kernel and a measure, respectively. Given a connected ordered graph $G$ on $n$ vertices, consider the multi-linear form $$ \Lambda_G(f_1,f_2, \dots, f_n)=\int_{x^1, \dots, x^n \in X} \ \prod_{(i,j) \in {\mathcal E}(G)} K(x^i,x^j) \prod_{l=1}^n f_l(x^l) d\mu(x^l), $$ where ${\mathcal E}(G)$ is the edge set of $G$. Define $\Lambda_G(p_1, \ldots, p_n)$ as the smallest constant $C>0$ such that the inequality $$ \Lambda_G(f_1, \dots, f_n) \leq C \prod_{i=1}^n {||f_i||}_{L^{p_i}(X, \mu)}$$ holds for all non-negative real-valued functions $f_i$, $1\le i\le n$, on $X$. The basic question is, how does the structure of $G$ and the mapping properties of the operator $T_K$ influence the sharp exponents. In this paper, this question is investigated mainly in the case $X={\Bbb F}_q^d$, the $d$-dimensional vector space over the field with $q$ elements, and $K(x^i,x^j)$ is the indicator function of the sphere evaluated at $x^i-x^j$. This provides a connection with the study of $L^p$-improving measures and distance set problems., Comment: 51 pages, 8 figures, typos fixed

Understanding the stability limitations and defect formation mechanisms in 2D magnets is essential for their utilization in spintronic and memory technologies. Here, we correlate defects in mono- to multilayer CrSBr with their structural, vibrational and magnetic properties. We use resonant Raman scattering to reveal distinct vibrational defect signatures. In pristine CrSBr, we show that bromine atoms mediate vibrational interlayer coupling, allowing for distinguishing between surface and bulk defect modes. We show that environmental exposure causes drastic degradation in monolayers, with the formation of intralayer defects. Through deliberate ion irradiation, we tune the formation of defect modes, which we show are strongly polarized and resonantly enhanced, reflecting the quasi-1D electronic character of CrSBr. Strikingly, we observe pronounced signatures of spin-phonon coupling of the intrinsic phonon modes and the ion beam induced defect modes throughout the magnetic transition temperature. Overall, we demonstrate that CrSBr shows air stability above the monolayer threshold, and provide further insight into the quasi-1D physics present. Moreover, we demonstrate defect engineering of magnetic properties and show that resonant Raman spectroscopy can serve as a direct fingerprint of magnetic phases and defects in CrSBr., Comment: main: 15 pages, 5 figures; SI: 16 pages, 15 figures

In this paper, we provide estimates for the additive discretized energy of \[\sum_{c\in C} |\{(a_1, a_2, b_1, b_2)\in A^2\times B^2: |(a_1 +cb_1) - (a_2 + cb_2)|\le \delta\}|_{\delta},\] that depend on non-concentration conditions of the sets. Our proof follows the Guth-Katz-Zahl approach (2021) with appropriate changes along the way clarifying and optimizing many of the steps. Several applications will also be discussed.

In this paper we obtain improved dimensional thresholds for dot product sets corresponding to compact subsets of a paraboloid. As a direct application of these estimates, we obtain significant improvements to the best known dimensional thresholds that guarantee that a given compact subset of Euclidean space determines a positive proportion of all possible congruence classes of simplexes. In many regimes this improves the results previously obtained by Erdogan-Hart-Iosevich (\cite{EHI}), Greenleaf-Iosevich-Liu-Palsson (\cite{GILP}) and others.

Let $V$ be a variety in $\mathbb{F}_q^d$ and $E\subset V$. It is known that if any line passing through the origin contains a bounded number of points from $E$, then $|\prod(E)|=|\{x\cdot y\colon x, y\in E\}|\gg q$ whenever $|E|\gg q^{\frac{d}{2}}$. In this paper, we show that the barrier $\frac{d}{2}$ can be broken when $V$ is a paraboloid in some specific dimensions. The main novelty in our approach is to link this question to the distance problem in one lower dimensional vector space, allowing us to use recent developments in this area to obtain improvements.

In 2017, by using the group theoretic approach, Bennett, Hart, Iosevich, Pakianathan, and Rudnev obtained a number of results on the distribution of simplices and sum-product type problems. The main purpose of this paper is to give a series of new applications of their powerful framework, namely, we focus on the product and quotient of distance sets, the $L^2$-norm of the direction set, and the $L^2$-norm of scales in difference sets., Comment: v2: accepted version

In this paper, we provide a general framework for counting geometric structures in pseudo-random graphs. As applications, our theorems recover and improve several results on the finite field analog of questions originally raised in the continuous setting. The results present interactions between discrete geometry, geometric measure theory, and graph theory.

While contextualized word embeddings have been a de-facto standard, learning contextualized phrase embeddings is less explored and being hindered by the lack of a human-annotated benchmark that tests machine understanding of phrase semantics given a context sentence or paragraph (instead of phrases alone). To fill this gap, we propose PiC -- a dataset of ~28K of noun phrases accompanied by their contextual Wikipedia pages and a suite of three tasks for training and evaluating phrase embeddings. Training on PiC improves ranking models' accuracy and remarkably pushes span-selection (SS) models (i.e., predicting the start and end index of the target phrase) near-human accuracy, which is 95% Exact Match (EM) on semantic search given a query phrase and a passage. Interestingly, we find evidence that such impressive performance is because the SS models learn to better capture the common meaning of a phrase regardless of its actual context. SotA models perform poorly in distinguishing two senses of the same phrase in two contexts (~60% EM) and in estimating the similarity between two different phrases in the same context (~70% EM)., Comment: Accepted to EACL 2023

Electric machine design optimization is a computationally expensive multi-objective optimization problem. While the objectives require time-consuming finite element analysis, optimization constraints can often be based on mathematical expressions, such as geometric constraints. This article investigates this optimization problem of mixed computationally expensive nature by proposing an optimization method incorporated into a popularly-used evolutionary multi-objective optimization algorithm - NSGA-II. The proposed method exploits the inexpensiveness of geometric constraints to generate feasible designs by using a custom repair operator. The proposed method also addresses the time-consuming objective functions by incorporating surrogate models for predicting machine performance. The article successfully establishes the superiority of the proposed method over the conventional optimization approach. This study clearly demonstrates how a complex engineering design can be optimized for multiple objectives and constraints requiring heterogeneous evaluation times and optimal solutions can be analyzed to select a single preferred solution and importantly harnessed to reveal vital design features common to optimal solutions as design principles.

Motivated by recent results on radial projections and applications to the celebrated Falconer distance problem, we study radial projections in the setting of finite fields. More precisely, we extend results due to Mattila and Orponen (2016), Orponen (2018), and Liu (2020) to finite spaces. In some cases, our results are stronger than the corresponding results in the continuous setting. In particular, we solve the finite field analog of a conjecture due to Liu and Orponen on the exceptional set of radial projections of a set of dimension between $d-2$ and $d-1$., Comment: V2: Proof of Lemma 3.1 is corrected

The main purpose of this paper is to provide threshold functions for the events that a random subset of the points of a finite vector space has certain properties related to point-flat incidences. Specifically, we consider the events that there is an $\ell$-rich $m$-flat with regard to a random set of points in $\mathbb{F}_q^n$, the event that a random set of points is an $m$-blocking set, and the event that there is an incidence between a random set of points and a random set of $m$-flats. One of our key ingredients is a stronger version of a recent result obtained by Chen and Greenhill (2021)., Comment: 17 pages

The recent breakthrough of Guth, Iosevich, Ou, and Wang (2019) on the Falconer distance problem states that for a compact set $A\subset \mathbb{R}^2$, if the Hausdorff dimension of $A$ is greater than $\frac{5}{4}$, then the distance set $\Delta(A)$ has positive Lebesgue measure. In a very recent paper, Murphy, Petridis, Pham, Rudnev, and Stevens (2022) proved the prime field version of this result, namely, for $E\subset\mathbb{F}_p^2$ with $|E|\gg p^{5/4}$, there exist many points $x\in E$ such that the number of distinct distances from $x$ is at least $cp$. The main purpose of this paper is to provide extensions in a very general structure of pinned trees, which is inspired by the recent work due to Ou and Taylor (2021).

Understanding structure at the interface between two-dimensional (2D) materials and 3D metals is crucial for designing novel 2D/3D heterostructures and improving the performance of many 2D material devices. Here, we quantify and discuss the 2D/3D interface structure and the 3D morphology in several materials systems. We first deposit facetted Au nanoislands on graphene and transition metal dichalcogenides, using measurements of the equilibrium island shape to determine values for the 2D/Au interface energies and examining the role of surface reconstructions, chemical identity, and defects on the grown structures. We then deposit the technologically relevant metals Ti and Nb under conditions where kinetic rather than thermodynamic factors govern growth. We describe a transition from dendritic to facetted islands as a function of growth temperature and discuss the factors determining island shape in these materials systems. Finally, we show that suspended 2D materials enable the fabrication of a novel type of 3D/2D/3D heterostructure and discuss the growth mechanism. We suggest that emerging nanodevices will utilize such versatile fabrication of 2D/3D heterostructures with well-characterized interfaces and morphologies.

Control of nucleation sites is an important goal in materials growth: nuclei in regular arrays may show emergent photonic or electronic behavior, and once the nuclei coalesce into thin films, the nucleation density influences parameters such as surface roughness, stress, and grain boundary structure. Tailoring substrate properties to control nucleation is therefore a powerful tool for designing functional thin films and nanomaterials. Here, we examine nucleation control for metals deposited on two-dimensional (2D) materials in a situation where substrate effects are absent and heterogeneous nucleation sites are minimized. Through quantification of faceted, epitaxial Au island nucleation on graphene, we show that ultra-low nucleation densities with nuclei several micrometers apart can be achieved on suspended graphene under conditions where we measure the nucleation density to be 2-3 orders of magnitude higher on the adjacent supported substrate. We estimate diffusion distances using nucleation theory and find a strong sensitivity of nucleation and diffusion to suspended graphene thickness. We suggest that nucleation site density control via substrate tuning may act as a platform for applications where non-lithographic patterning of metals on graphene is beneficial.

Let $\mathbb{F}_q$ be a finite field of order $q$. Iosevich and Rudnev (2005) proved that for any set $A\subset \mathbb{F}_q^d$, if $|A|\gg q^{\frac{d+1}{2}}$, then the distance set $\Delta(A)$ contains a positive proportion of all distances. Although this result is sharp in odd dimensions, it is conjectured that the right exponent should be $\frac{d}{2}$ in even dimensions. During the last 15 years, only some improvements have been made in two dimensions, and the conjecture is still wide open in higher dimensions. To fill the gap, we need to understand more about the structures of the distance sets, the main purpose of this paper is to provide some structural theorems on the distribution of square and non-square distances., Comment: V2 includes several fixes and clarifications

In this paper, we prove the first incidence bound for points and conics over prime fields. As applications, we prove new results on expansion of bivariate polynomial images and on certain variations of distinct distances problems. These include new lower bounds on the number of pinned algebraic distances as well as improvements of results of Koh and Sun (2014) and Shparlinski (2006) on the size of the distance set formed by two large subsets of finite dimensional vector spaces over finite fields. We also prove a variant of Beck's theorem for conics., Comment: To appear in European Journal of Combinatorics

A principle behind dozens of attribution methods is to take the prediction difference between before-and-after an input feature (here, a token) is removed as its attribution. A popular Input Marginalization (IM) method (Kim et al., 2020) uses BERT to replace a token, yielding more plausible counterfactuals. While Kim et al. (2020) reported that IM is effective, we find this conclusion not convincing as the DeletionBERT metric used in their paper is biased towards IM. Importantly, this bias exists in Deletion-based metrics, including Insertion, Sufficiency, and Comprehensiveness. Furthermore, our rigorous evaluation using 6 metrics and 3 datasets finds no evidence that IM is better than a Leave-One-Out (LOO) baseline. We find two reasons why IM is not better than LOO: (1) deleting a single word from the input only marginally reduces a classifier's accuracy; and (2) a highly predictable word is always given near-zero attribution, regardless of its true importance to the classifier. In contrast, making LIME samples more natural via BERT consistently improves LIME accuracy under several ROAR metrics., Comment: 9 pages. Long paper to appear at AACL-IJCNLP 2022