共找到 20 条结果
Common subword tokenization algorithms like BPE and UnigramLM assume that text can be split into meaningful units by concatenative measures alone. This is not true for languages such as Hebrew and Arabic, where morphology is encoded in root-template patterns, or Malay and Georgian, where split affixes are common. We present SPLINTER, a pre-processing step which rearranges text into a linear form that better represents such nonconcatenative morphologies, enabling meaningful contiguous segments to be found by the tokenizer. We demonstrate SPLINTER's merit using both intrinsic measures evaluating token vocabularies in Hebrew, Arabic, and Malay; as well as on downstream tasks using BERT-architecture models trained for Hebrew.
Once united in support of President Trump, right-wing influencers are embroiled in feuds and losing viewers, creating a problem for Republicans in the midterm elections
Performing computations while maintaining privacy is an important problem in todays distributed machine learning solutions. Consider the following two set ups between a client and a server, where in setup i) the client has a public data vector $\mathbf{x}$, the server has a large private database of data vectors $\mathcal{B}$ and the client wants to find the inner products $\langle \mathbf{x,y_k} \rangle, \forall \mathbf{y_k} \in \mathcal{B}$. The client does not want the server to learn $\mathbf{x}$ while the server does not want the client to learn the records in its database. This is in contrast to another setup ii) where the client would like to perform an operation solely on its data, such as computation of a matrix inverse on its data matrix $\mathbf{M}$, but would like to use the superior computing ability of the server to do so without having to leak $\mathbf{M}$ to the server. \par We present a stochastic scheme for splitting the client data into privatized shares that are transmitted to the server in such settings. The server performs the requested operations on these shares instead of on the raw client data at the server. The obtained intermediate results are sent back t
We formulate a lattice Hamiltonian model for intermediate energy heavy ion collisions. Our approach incorporates a momentum-dependent nuclear mean field that yields an optical potential that agrees with proton-nucleus experiments and that also receives support from measurements involving heavy ions. We emphasize the precision of our numerical results in connection with energy and momentum conservation. We first outline the static properties of our solutions, after which we consider recent stopping power results for heavy ion collisions. The results obtained with different types of nuclear mean fields are compared with each other. Consequences of the stopping power data on the nuclear equation of state are outlined.
A convex splinter $K$ is a union of convex sets in $\mathbb{R}^d$ such that every minimal affine dependent set of $\mathbb{R}^d$ contained in $K$ is contained in one of the sets. The study of convex splinters was motivated by the study of flat transversals to convex sets. We extend several variations of Helly's theorem from convex geometry to convex splinters. These include fractional and colorful variations of Helly's theorem. We also extend Tverberg's theorem to convex splinters.
We introduce (lim-)perfectoid splitting, which is a global variant of (lim-)perfectoid purity. Our main result establishes a correspondence between the lim-perfectoid splitting of projective schemes and the lim-perfectoid purity of their Gorenstein section rings. As an application, we construct a new supply of examples of lim-perfectoid pure rings that go beyond the previously known complete intersection or splinter-type cases.
Search engines have become the gateway to information, products, and services, including those concerning healthcare. Access to reproductive health has been especially complicated in the wake of the 2022 Dobbs v. Jackson decision by the Supreme Court of the United States, splintering abortion regulations among the states. In this study, we performed an audit of the advertisements shown to Google Search users seeking information about abortion across the United States during the year following the Dobbs decision. We found that Crisis Pregnancy Centers (CPCs) -- organizations that target women with unexpected or "crisis" pregnancies, but do not provide abortions -- accounted for 47% of advertisements, whereas abortion clinics -- for 30%. Advertisements from CPCs were often returned for queries concerning information and safety. The type of advertisements returned, however, varied widely within each state, with Arizona returning the most advertisements from abortion clinics and other pro-choice organizations, and Minnesota the least. The proportion of pro-choice vs. anti-choice advertisements returned also varied over time, but estimates from Staggered Augmented Synthetic Control Meth
This work is concerned with categorical methods for studying singularities. Our focus is on birational derived splinters, which is a notion that extends the definition of rational singularities beyond varieties over fields of characteristic zero. Particularly, we show that an invariant called `level' in the associated derived category measures the failure of these singularities.
Swarmalators are active agents that move in position space and exhibit internal degrees of freedom. Due to interactions of their positions and phases of oscillation, they show on the one hand swarming, similar to the effect of flocking of birds. In addition, they exhibit synchronization behavior, analogous to what has been observed in fireflies. Previous works studied scenarios in which the phases are forced externally. Here, we consider a pseudo-force that acts on the positions of the swarmalators. Due to the resulting attraction towards the center of position space, transitions from the splintered and active phase-wave state to the static antisynchronized state are found. To quantify the crystal order of swarmalators, we introduce an order parameter that is based on the Fourier transform of their positions.
We prove that a Noetherian ring $R$ is a splinter if and only if for every equidimensional surjective morphism $\operatorname{Spec}(S) \to \operatorname{Spec}(R)$, the map $R \to S$ is pure. This yields a large, nontrivial class of ring maps that are automatically pure. More generally, we prove that a locally Noetherian scheme $Y$ is locally a splinter if and only if every locally equidimensional morphism $X \to Y$ is strongly pure. Special cases of our results show that equidimensional fibrations over normal $\mathbf{Q}$-schemes or regular schemes of arbitrary characteristic are strongly pure. The main ingredient is a new factorization result for locally equidimensional morphisms of schemes, which is of independent interest. Additionally, we prove a weak Boutot-type theorem for $F$-rationality, which says that $F$-rationality descends under pure ring maps that are locally equidimensional under universally catenary assumptions. This statement is false without the locally equidimensional hypothesis.
If $(R,\mathfrak{m})$ is a complete local ring of mixed characteristic $(0,p)$ and $R/pR$ is an $F$-pure Gorenstein domain, we find a sufficient condition for $R$ to be perfectoid pure. This condition is related to the Cohen-Macaulayness of the absolute integral closures of Gorenstein local domains of mixed characteristic which are not necessarily excellent. Along the way, we show that the problem of lifting $F$-purity of $R/pR$ to perfectoid purity of $R$ is equivalent to a similar deformation problem for the splinter property.
This article is a primer on concept extrapolation - the ability to take a concept, a feature, or a goal that is defined in one context and extrapolate it safely to a more general context. Concept extrapolation aims to solve model splintering - a ubiquitous occurrence wherein the features or concepts shift as the world changes over time. Through discussing value splintering and value extrapolation the article argues that concept extrapolation is necessary for Artificial Intelligence alignment.
Given a right adjoint functor between triangulated categories and an object in the target category, we show that the unit map of adjunction on that object is a split monomorphism if and only if the object belongs to the additive closure of (all possible) shifts of an object in the image of the functor. Applications to geometric context related to (derived) splinters and rational singularities are given.
We construct a local Noetherian splinter (in fact, a weakly $F$-regular domain) in prime characteristic which is not catenary, which we view as an analogue of a theorem of Ogoma in equal characteristic zero. Moreover, we construct a weakly $F$-regular local UFD which is not Cohen-Macaulay. Both of these examples are obtained via finding sufficient conditions ensuring that a complete local ring of prime characteristic is the completion of some weakly $F$-regular local domain, which we expect to be of independent interest.
This work presents a range of triangulated characterizations for important classes of singularities such as derived splinters, rational singularities, and Du Bois singularities. An invariant called 'level' in a triangulated category can be used to measure the failure of a variety to have a prescribed singularity type. We provide explicit computations of this invariant for reduced Nagata schemes of Krull dimension one and for affine cones over smooth projective hypersurfaces. Furthermore, these computations are utilized to produce upper bounds for Rouquier dimension on the respective bounded derived categories.
While the splinter property is a local property for Noetherian schemes in characteristic zero, Bhatt observed that it imposes strong conditions on the global geometry of proper schemes in positive characteristic. We show that if a proper scheme over a field of positive characteristic is a splinter, then its Nori fundamental group scheme is trivial and its Kodaira dimension is negative. In another direction, Bhatt also showed that any splinter in positive characteristic is a derived splinter. We ask whether the splinter property is a derived invariant for projective varieties in positive characteristic and give a positive answer for normal Gorenstein projective varieties with big anticanonical divisor. We also show that global F-regularity is a derived invariant for normal Gorenstein projective varieties in positive characteristic.
We show that certain ways of solving some combinatorial optimization problems can be understood as using query planes to divide the space of problem instances into polyhedra that could fit into those that characterize the problem's various solutions. This viewpoint naturally leads to a splinter-proneness property that is then shown to be responsible for the hardness of the concerned problem. We conjecture that the $NP$-equivalent traveling salesman problem (TSP) has this property and hence is hard to solve to a certain extent.
Let $X$ be a normal, excellent, noetherian scheme over $\operatorname{Spec}\mathbb{Q}$ with a dualizing complex. In this note, we find an alternate characterization of the multiplier ideal of $X$, as defined by de Fernex-Hacon, by considering maps $π_*ω_Y\to\mathcal{O}_X$ where $π:Y\to X$ ranges over all regular alterations. As a corollary to this result, we give a derived splinter characterization of klt singularities, akin to the characterization of rational singularities given by Kovács and Bhatt. We also give an analogous description of the test ideal in characteristic $p>2$ as a corollary to a result of Epstein-Schwede.
A Noetherian reduced ring $A$ is called a birational derived splinter if for all proper birational maps $X\to\operatorname{Spec}(A)$, the canonical map $A\to Rf_*\mathcal{O}_X$ splits. In equal characteristic zero this property characterizes rational singularities, but much less can be said in positive or mixed characteristics. In this paper, we prove some fundamental properties of this notion, including the behavior under localization, taking a pure subring, taking direct limit, and along an étale extension. In particular, direct limit of rational singularities in characteristic zero has rational singularities. Then, we study residue extensions (in arbitrary characteristic), and openness and regular extensions in positive characteristic, parallel to Datta-Tucker and the author's previous works on splinters.
Modern GPUs face a trade-off on how the page size used for memory management affects address translation and demand paging. Support for multiple page sizes can help relax the page size trade-off so that address translation and demand paging optimizations work together synergistically. However, existing page coalescing and splintering policies require costly base page migrations that undermine the benefits multiple page sizes provide. In this paper, we observe that GPGPU applications present an opportunity to support multiple page sizes without costly data migration, as the applications perform most of their memory allocation en masse (i.e., they allocate a large number of base pages at once). We show that this en masse allocation allows us to create intelligent memory allocation policies which ensure that base pages that are contiguous in virtual memory are allocated to contiguous physical memory pages. As a result, coalescing and splintering operations no longer need to migrate base pages. We introduce Mosaic, a GPU memory manager that provides application-transparent support for multiple page sizes. Mosaic uses base pages to transfer data over the system I/O bus, and allocates ph