A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
StrokeCoder uses Transformers to generate images from single examples.
New method defends against both single-step and iterative adversarial examples.
Due to the surprisingly good representation power of complex distributions, neural network (NN) classifiers are widely used in many tasks which include natural language processing, computer vision and cyber security. In recent works, people noticed the existence of adversarial examples. These adversarial examples break…
Single gradient step finds adversarial examples in random neural networks.
A new method reduces adversarial training time without overfitting.
New bounds show limitations of sample-wise information-theoretic generalization.
In recent years, the electrocardiogram (ECG) has seen a large diffusion in both medical and commercial applications, fueled by the rise of single-lead versions. Single-lead ECG can be embedded in medical devices and wearable products such as the injectable Medtronic Linq monitor, the iRhythm Ziopatch wearable monitor, …
New example of non-smooth isotopy after stabilization in 4-manifolds.
Paper tackles catastrophic overfitting in single-step adversarial training.
TOAD-GAN generates coherent game levels from a single example.
Adversarial examples are malicious inputs designed to fool machine learning models. They often transfer from one model to another, allowing attackers to mount black box attacks without knowledge of the target model's parameters. Adversarial training is the process of explicitly training a model on adversarial examples,…
We study the interplay between memorization and generalization of overparameterized networks in the extreme case of a single training example and an identity-mapping task. We examine fully-connected and convolutional networks (FCN and CNN), both linear and nonlinear, initialized randomly and then trained to minimize th…
Lazy neural networks are vulnerable to adversarial attacks.
New examples of Lie algebras with ad-invariant metrics found.
Multi-head attention outperforms single-head in in-context linear regression tasks.
A gap in a paper of Rubinstein-Scharlemann is explored: new examples are found of closed orientable 3-manifolds with possibly multiple genus 2 Heegaard splittings. Properties common to all the examples in the original paper are not universally shared by the new examples: some of the new examples have Hempel distance 3,…
This paper proposes multi-view attack strategies for deep models.
The family of embedded, singly periodic minimal surfaces of Riemann have as limit-surfaces the helicoid, the catenoid, a single plane, or an infinite set of equally-spaced parallel planes.
A complementary label (CL) simply indicates an incorrect class of an example, but learning with CLs results in multi-class classifiers that can predict the correct class. Unfortunately, the problem setting only allows a single CL for each example, which notably limits its potential since our labelers may easily identif…
New method certifies joint adversarial robustness of model ensembles.
Simultaneously recorded electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) can be used to non-invasively measure the spatiotemporal dynamics of the human brain. One challenge is dealing with the artifacts that each modality introduces into the other when the two are recorded concurrently, for…
We use the momentum construction of Calabi to study the conical Kähler-Ricci flow on Hirzebruch surfaces with cone angle along the exceptional curve, and show that either the flow Gromov-Hausdorff converges to the Riemann sphere or a single point in finite time, or the flow contracts the cone divisor to a single point …
Neural networks struggle with extrapolation, but a new framework allows them to learn counterfactual invariances.
This paper explores the use of adversarial examples in training speech recognition systems to increase robustness of deep neural network acoustic models. During training, the fast gradient sign method is used to generate adversarial examples augmenting the original training data. Different from conventional data augmen…
We extend the common Poisson shock framework reviewed for example in Lindskog and McNeil (2003) to a formulation avoiding repeated defaults, thus obtaining a model that can account consistently for single name default dynamics, cluster default dynamics and default counting process. This approach allows one to introduce…
We provide in this note two relevant examples of Lagrangian cobordisms. The first one gives an example of two exact Lagrangian submanifolds which cannot be composed in an exact fashion. The second one is an example of an exact Lagrangian cobordism on which all primitive of the Liouville form is not constant on the nega…
By studying the example of smooth structures on CP^2#3(-CP^2) we illustrate how surgery on a single embedded nullhomologous torus can be utilized to change the symplectic structure, the Seiberg-Witten invariant, and hence the smooth structure on a 4-manifold.
Paper introduces ICGNs to model convex gradients.
Casson and Gordon gave the rectangle condition for strong irreducibility of Heegaard splittings [1]. We give a parity condition for irreducibility of Heegaard splittings of irreducible manifolds. As an application, we give examples of non-stabilized Heegaard splittings by doing a single Dehn twist.
Adversarial examples are slight perturbations that are designed to fool artificial neural networks when fed as an input. In this work the usability of the Fisher information for the detection of such adversarial attacks is studied. We discuss various quantities whose computation scales well with the network size, study…
Extending an example by Colding and Minicozzi, we construct a sequence of properly embedded minimal disks in an infinite Euclidean cylinder around the -axis with curvature blow-up at a single point. The sequence converges to a non smooth and non proper minimal lamination in the cylinder. Moreover, we show th…
Found first example of homogeneous gradient solitons for G-Laplacian flow.
Inspired by the phenomenon of catastrophic forgetting, we investigate the learning dynamics of neural networks as they train on single classification tasks. Our goal is to understand whether a related phenomenon occurs when data does not undergo a clear distributional shift. We define a `forgetting event' to have occur…
Mathematical properties of the historical GDP/cap distributions are discussed and explained. These distributions are frequently incorrectly interpreted and the Unified Growth Theory is an outstanding example of such common misconceptions. It is shown here that the fundamental postulates of this theory are contradicted …
In distributed statistical learning, samples are split across machines and a learner wishes to use minimal communication to learn as well as if the examples were on a single machine. This model has received substantial interest in machine learning due to its scalability and potential for parallel speedup. Howev…
Kuperberg [Algebr. Geom. Topol. 3 (2003) 587-591] has shown that a virtual knot corresponds (up to generalized Reidemeister moves) to a unique embedding in a thichened surface of minimal genus. If a virtual knot diagram is equivalent to a classical knot diagram then this minimal surface is a sphere. Using this result a…
Given knots K and J, one can ask whether a single smoothing of a crossing in a diagram for K can convert it into a diagram for J. As an interesting example, Zekovic discovered that the torus knot T(2,5) can be converted into T(2,-5) with a single smoothing. On the other hand, Moore and Vasquez have shown that among tor…
A fast single-shot MC dropout method for neural networks.
We provide single-model estimates of aleatoric and epistemic uncertainty for deep neural networks. To estimate aleatoric uncertainty, we propose Simultaneous Quantile Regression (SQR), a loss function to learn all the conditional quantiles of a given target variable. These quantiles can be used to compute well-calibrat…
New offline RL method handles average-reward MDPs with single-policy coverage.
Study on indexability of restless multi-armed bandits and rollout policy performance.
Many deep learning algorithms can be easily fooled with simple adversarial examples. To address the limitations of existing defenses, we devised a probabilistic framework that can generate an exponentially large ensemble of models from a single model with just a linear cost. This framework takes advantage of neural net…
The paper finds determinantal expressions for certain symmetric space integrals.
We provide the asymptotic distribution of the major indexes used in the statistical literature to quantify disparate treatment in machine learning. We aim at promoting the use of confidence intervals when testing the so-called group disparate impact. We illustrate on some examples the importance of using confidence int…
In this paper we prove that the limit set of any Weil-Petersson geodesic ray with uniquely ergodic ending lamination is a single point in the Thurston compactification of Teichmüller space. On the other hand, we construct examples of Weil-Petersson geodesics with minimal nonuniquely ergodic ending laminations and limit…
Single neuron sensitivity reveals model weaknesses.
We develop a way of seeing a complete orientable hyperbolic -manifold as an orbifold cover of a Coxeter polytope that has a facet colouring. We also develop a way of finding totally geodesic sub-manifolds in , and describing the result of mu…
A depth surface of E^3 is a range image observed from a single view can be represented by a digital graph (Monge patch) surface . That is, a depth or range value at a point (u,v) is given by a single valued function z=f(u,v). In the present study we consider the surfaces in Euclidean 4-space E^4 given with a Monge patc…