Search

Spectral Embedding (SE) is a popular method for dimensionality reduction, applicable across diverse domains. Nevertheless, its current implementations face three prominent drawbacks which curtail its broader applicability: generalizability (i.e., out-of-sample extension), scalability, and eigenvectors separation. In this paper, we introduce GrEASE: Generalizable and Efficient Approximate Spectral Embedding, a novel deep-learning approach designed to address these limitations. GrEASE incorporates an efficient post-processing step to achieve eigenvectors separation, while ensuring both generalizability and scalability, allowing for the computation of the Laplacian's eigenvectors on unseen data. This method expands the applicability of SE to a wider range of tasks and can enhance its performance in existing applications. We empirically demonstrate GrEASE's ability to consistently approximate and generalize SE, while ensuring scalability. Additionally, we show how GrEASE can be leveraged to enhance existing methods. Specifically, we focus on UMAP, a leading visualization technique, and introduce NUMAP, a generalizable version of UMAP powered by GrEASE. Our codes are publicly available.

We prove that the Kazhdan-Lusztig basis of Specht modules is upper triangular with respect to all generalized Gelfand-Tsetlin bases constructed from any multiplicity-free tower of standard parabolic subgroups., Comment: 10 pages, extended abstract for submission to FPSAC 2025

Multi-view representation learning (MvRL) has garnered substantial attention in recent years, driven by the increasing demand for applications that can effectively process and analyze data from multiple sources. In this context, graph Laplacian-based MvRL methods have demonstrated remarkable success in representing multi-view data. However, these methods often struggle with generalization to new data and face challenges with scalability. Moreover, in many practical scenarios, multi-view data is contaminated by noise or outliers. In such cases, modern deep-learning-based MvRL approaches that rely on alignment or contrastive objectives present degraded performance in downstream tasks, as they may impose incorrect consistency between clear and corrupted data sources. We introduce $\textit{SpecRaGE}$, a novel fusion-based framework that integrates the strengths of graph Laplacian methods with the power of deep learning to overcome these challenges. SpecRage uses neural networks to learn parametric mapping that approximates a joint diagonalization of graph Laplacians. This solution bypasses the need for alignment while enabling generalizable and scalable learning of informative and meaningful representations. Moreover, it incorporates a meta-learning fusion module that dynamically adapts to data quality, ensuring robustness against outliers and noisy views. Our extensive experiments demonstrate that SpecRaGE outperforms state-of-the-art methods, particularly in scenarios with data contamination, paving the way for more reliable and efficient multi-view learning.

We introduce a new algorithm for finding kernel elements in the Burau representation. Our algorithm applies reservoir sampling to a statistic on matrices which is closely correlated with Garside length. Using this we exhibit an explicit kernel element in the Burau representation on 4-strands reduced modulo 5., Comment: 10 pages, 5 figures

We study representations of simply-laced Weyl groups which are equipped with canonical bases. Our main result is that for a large class of representations, the separable elements of the Weyl group $W$ act on these canonical bases by bijections up to lower-order terms. Examples of this phenomenon include the action of separable permutations on the Kazhdan--Lusztig basis of irreducible representations for the symmetric group, and the action of separable elements of $W$ on dual canonical bases of weight zero in tensor product representations of a Lie algebra. Our methods arise from categorical representation theory, and in particular the study of the perversity of Rickard complexes acting on triangulated categories., Comment: 32 pages

We develop a theory of parabolic induction and restriction functors relating modules over Coulomb branch algebras, in the sense of Braverman-Finkelberg-Nakajima. Our functors generalize Bezrukavnikov-Etingof's induction and restriction functors for Cherednik algebras, but their definition uses different tools. After this general definition, we focus on quiver gauge theories attached to a quiver $\Gamma$. The induction and restriction functors allow us to define a categorical action of the corresponding symmetric Kac-Moody algebra $\mathfrak{g}_{\Gamma}$ on category $ \mathcal O $ for these Coulomb branch algebras. When $ \Gamma $ is of Dynkin type, the Coulomb branch algebras are truncated shifted Yangians and quantize generalized affine Grassmannian slices. Thus, we regard our action as a categorification of the geometric Satake correspondence. To establish this categorical action, we define a new class of "flavoured" KLRW algebras, which are similar to the diagrammatic algebras originally constructed by the second author for the purpose of tensor product categorification. We prove an equivalence between the category of Gelfand-Tsetlin modules over a Coulomb branch algebra and the modules over a flavoured KLRW algebra. This equivalence relates the categorical action by induction and restriction functors to the usual categorical action on modules over a KLRW algebra., Comment: 66 pages, version 2: many corrections, improved treatment of GK dimension, 71 pages

This paper addresses the problem of defect segmentation in semiconductor manufacturing. The input of our segmentation is a scanning-electron-microscopy (SEM) image of the candidate defect region. We train a U-net shape network to segment defects using a dataset of clean background images. The samples of the training phase are produced automatically such that no manual labeling is required. To enrich the dataset of clean background samples, we apply defect implant augmentation. To that end, we apply a copy-and-paste of a random image patch in the clean specimen. To improve the robustness of the unlabeled data scenario, we train the features of the network with unsupervised learning methods and loss functions. Our experiments show that we succeed to segment real defects with high quality, even though our dataset contains no defect examples. Our approach performs accurately also on the problem of supervised and labeled defect segmentation.

The complex irreducible representations of the symmetric group carry an important canonical basis called the Kazhdan-Lusztig basis. Although it is difficult to express how general permutations act on this basis, some distinguished permutations have beautiful descriptions. In 2010 Rhoades showed that the long cycle $(1, 2,..., n)$ acts by the jeu-de-taquin promotion operator in the case when the irreducible representation is indexed by a rectangular partition. We prove a generalisation of this theorem in two directions: on the one hand we lift the restriction on the shape of the partition, and on the other hand we enlarge the result to the collection of all separable permutations., Comment: 12 pages, extended abstract to appear in proceedings of FPSAC 2022

The cactus group acts on the set of standard Young tableau of a given shape by (partial) Sch\"utzenberger involutions. It is natural to extend this action to the corresponding Specht module by identifying standard Young tableau with the Kazhdan-Lusztig basis. We term these representations of the cactus group "Sch\"utzenberger modules", denoted $S^\lambda_{\mathsf{Sch}}$, and in this paper we investigate their decomposition into irreducible components. We prove that when $\lambda$ is a hook shape, the cactus group action on $S^\lambda_{\mathsf{Sch}}$ factors through $S_{n-1}$ and the resulting multiplicities are given by Kostka coefficients. Our proof relies on results of Berenstein and Kirillov and Chmutov, Glick, and Pylyavskyy., Comment: v2: 15 pages, final version to appear in Algebraic Combinatorics

Let $\mathfrak{g}$ be a semisimple simply-laced Lie algebra of finite type. Let $\mathcal{C}$ be an abelian categorical representation of the quantum group $U_q(\mathfrak{g})$ categorifying an integrable representation $V$. The Artin braid group $B$ of $\mathfrak{g}$ acts on $D^b(\mathcal{C})$ by Rickard complexes, providing a triangulated equivalence $\Theta_{w_0}:D^b(\mathcal{C}_\mu) \to D^b(\mathcal{C}_{w_0(\mu)})$, where $\mu$ is a weight of $V$ and $\Theta_{w_0}$ is a positive lift of the longest element of the Weyl group. We prove that this equivalence is t-exact up to shift when $V$ is isotypic, generalising a fundamental result of Chuang and Rouquier in the case $\mathfrak{g}=\mathfrak{sl}_2$. For general $V$, we prove that $\Theta_{w_0}$ is a perverse equivalence with respect to a Jordan-H\"older filtration of $\mathcal{C}$. Using these results we construct, from the action of $B$ on $V$, an action of the cactus group on the crystal of $V$. This recovers the cactus group action on $V$ defined via generalised Sch\"utzenberger involutions, and provides a new connection between categorical representation theory and crystal bases. We also use these results to give new proofs of theorems of Berenstein-Zelevinsky, Rhoades, and Stembridge regarding the action of symmetric group on the Kazhdan-Lusztig basis of its Specht modules., Comment: v3: 26 pages, fixed a mistake in Theorem 3.6. v2: 26 pages, revision based on referee comments. v1: 23 pages

In their study of local models of Shimura varieties for totally ramified extensions, Pappas and Rapoport posed a conjecture about the reducedness of a certain subscheme of $n \times n$ matrices. We give a positive answer to their conjecture in full generality. Our main ideas follow naturally from two of our previous works. The first is our proof of a conjecture of Kreiman, Lakshmibai, Magyar, and Weyman on the equations defining type A affine Grassmannians. The second is the work of the first two authors and Kamnitzer on affine Grassmannian slices and their reduced scheme structure. We also present a version of our argument that is almost completely elementary: the only non-elementary ingredient is the Frobenius splitting of Schubert varieties., Comment: 10 pages, comments welcome

Truncated shifted Yangians are a family of algebras which naturally quantize slices in the affine Grassmannian. These algebras depend on a choice of two weights $\lambda$ and $\mu$ for a Lie algebra $\mathfrak{g}$, which we will assume is simply-laced. In this paper, we relate the category $\mathcal{O}$ over truncated shifted Yangians to categorified tensor products: for a generic integral choice of parameters, category $\mathcal{O}$ is equivalent to a weight space in the categorification of a tensor product of fundamental representations defined by the third author using KLRW algebras. We also give a precise description of category $\mathcal{O}$ for arbitrary parameters using a new algebra which we call the parity KLRW algebra. In particular, we confirm the conjecture of the authors that the highest weights of category $\mathcal{O}$ are in canonical bijection with a product monomial crystal depending on the choice of parameters. This work also has interesting applications to classical representation theory. In particular, it allows us to give a classification of simple Gelfand-Tsetlin modules of $U(\mathfrak{gl}_n)$ and its associated W-algebras., Comment: v3: final version for publication in Proceedings of the LMS

The affine Grassmannian of $SL_n$ admits an embedding into the Sato Grassmannian, which further admits a Pl\"ucker embedding into the projectivization of Fermion Fock space. Kreiman, Lakshmibai, Magyar, and Weyman describe the linear part of the ideal defining this embedding in terms of certain elements of the dual of Fock space called "shuffles", and they conjecture that these elements together with the Pl\"ucker relations suffice to cut out the affine Grassmannian. We give a proof of this conjecture in two steps: first we reinterpret the shuffles equations in terms of Frobenius twists of symmetric functions. Using this, we reduce to a finite dimensional-problem, which we solve. For the second step we introduce a finite-dimensional analogue of the affine Grassmannians of $SL_n$, which we conjecture to be precisely the reduced subscheme of a finite-dimensional Grassmannian consisting of subspaces invariant under a nilpotent operator., Comment: 33 pages, comments welcome; (v2) corrected typos, revised Sections 1.1, 3.1.5, 4.3 and argument in proof of Theorem 5.10, results unchanged

We present a quantization of an isomorphism of Mirkovi\'c and Vybornov which relates the intersection of a Slodowy slice and a nilpotent orbit closure in $\mathfrak{gl}_N$ , to a slice between spherical Schubert varieties in the affine Grassmannian of $PGL_n$ (with weights encoded by the Jordan types of the nilpotent orbits). A quantization of the former variety is provided by a parabolic W-algebra and of the latter by a truncated shifted Yangian. Building on earlier work of Brundan and Kleshchev, we define an explicit isomorphism between these non-commutative algebras, and show that its classical limit is a variation of the original isomorphism of Mirkovi\'c and Vybornov. As a corollary, we deduce that the W-algebra is free as a left (or right) module over its Gelfand-Tsetlin subalgebra, as conjectured by Futorny, Molev, and Ovsienko., Comment: v2: 48 pages. Major rewrite following referee comments. Added proof of a conjecture of Futorny, Molev, and Ovsienko that the finite W-algebra is free over its Gelfand-Tsetlin subalgebra

We introduce a simple diagrammatic 2-category $\mathscr{A}$ that categorifies the image of the Fock space representation of the Heisenberg algebra and the basic representation of $\mathfrak{sl}_\infty$. We show that $\mathscr{A}$ is equivalent to a truncation of the Khovanov--Lauda categorified quantum group $\mathscr{U}$ of type $A_\infty$, and also to a truncation of Khovanov's Heisenberg 2-category $\mathscr{H}$. This equivalence is a categorification of the principal realization of the basic representation of $\mathfrak{sl}_\infty$. As a result of the categorical equivalences described above, certain actions of $\mathscr{H}$ induce actions of $\mathscr{U}$, and vice versa. In particular, we obtain an explicit action of $\mathscr{U}$ on representations of symmetric groups. We also explicitly compute the Grothendieck group of the truncation of $\mathscr{H}$. The 2-category $\mathscr{A}$ can be viewed as a graphical calculus describing the functors of $i$-induction and $i$-restriction for symmetric groups, together with the natural transformations between their compositions. The resulting computational tool is used to give simple diagrammatic proofs of (apparently new) representation theoretic identities., Comment: 36 pages. v2: Minor corrections, published version

We prove in type A a conjecture which describes the ideal of transversal slices to spherical Schubert varieties in the affine Grassmannian. As a corollary, we prove a modular description (due to Finkelberg-Mirkovi\'c) of the spherical Schubert varieties., Comment: 12 pages, comments welcome

Truncated shifted Yangians are a family of algebras which are natural quantizations of slices in the affine Grassmannian. We study the highest weight representations of these algebras. In particular, we conjecture that the possible highest weights for these algebras are described by product monomial crystals, certain natural subcrystals of Nakajima's monomials. We prove this conjecture in type A. We also place our results in the context of symplectic duality and prove a conjecture of Hikita in this situation., Comment: 57 pages

The IceCube telescope has detected diffuse neutrino emission, 20 events of which were reported to be above 60~TeV. In this paper, we fit the diffuse neutrino spectrum using Poisson statistics, which are the most appropriate for the low counts per energy bin. We extend the fitted energy range and exploit the fact that no neutrinos were detected above 2~PeV, despite the high detector sensitivity around the Glashow resonance at 6.3\,PeV and beyond. A best-fit power-law slope of $\alpha=2.9\pm 0.3$ is found with no evidence for a high-energy cutoff. This slope is steeper than $\alpha=2.3\pm 0.3$ found by the IceCube team using a different fitting method. Such a steep spectrum facilitates the identification of high energy ($\gg$ PeV) neutrinos, if detected, to be due to the GZK effect of cosmic-ray protons interacting with the Extragalactic Background Light. We use the ratio of EeV to PeV neutrinos in GZK models to show that the currently detected PeV neutrinos could not be due to the GZK effect, because this would imply many more higher-energy neutrinos that should have been detected, but were not. The non-detection of GZK neutrinos by IceCube despite more than essentially 1200 observing days, has already ruled out (at 95\% confidence) models that predict rates of $\sim1$ neutrino/yr or more. We use this non-detection to quantify the confidence at which GZK models are ruled out, and compute the additional IceCube and (in the future) ARA observing time that would rule them out with 95\% confidence if no detection is made., Comment: 6 pages, 4 figures. Submitted to Astrophysical Journal

We construct a category of quantum polynomial functors which deforms Friedlander and Suslin's category of strict polynomial functors. The main aim of this paper is to develop from first principles the basic structural properties of this category (duality, projective generators, braiding etc.) in analogy with classical strict polynomial functors. We then apply the work of Hashimoto and Hayashi in this context to construct quantum Schur/Weyl functors, and use this to provide new and easy derivations of quantum $(GL_m,GL_n)$ duality, along with other results in quantum invariant theory., Comment: 34 pages, final version to appear in Journal of Algebra

The IceCube high-energy neutrino telescope has been collecting data since 2006. Conversely, hundreds of Gamma Ray Bursts (GRBs) have been detected by the GBM on board Fermi, since its launch in 2008. So far no neutrino event has been associated with a GRB, despite many models predicting the generation of high energy neutrinos through GRB photon interaction with PeV protons in the GRB jet. We use the non-detection of neutrinos to constrain the hadronic content of GRB jets independent of jet model parameters. Assuming a generic particle spectrum of $E^{-\alpha}$ with $\alpha = 2$, we find that the ratio of the energy carried by pions to that in electrons has to be small $f_\pi / f_e \lesssim 0.24$ at 95\% confidence level. A distribution of spectral slopes can lower $f_\pi / f_e$ by orders of magnitude. Another limit, independent of neutrinos, is obtained if one ascribes the measured Fermi/LAT GeV gamma-ray emission to pair-photon cascades of high-energy photons resulting from (the same photon-hadronic interactions and subsequent) neutral pion decays. Based on the generally observed MeV to GeV GRB fluence ratio of $\approx 10$, we show that $f_\pi / f_e \lesssim 0.3$. In some bursts, where this ratio is as low as unity, $f_\pi / f_e \lesssim 0.03$. These findings add to the mounting doubts regarding the presence of PeV protons in GRB jets., Comment: 6 pages, 5 figures. Submitted to Astrophysical Journal