Published on 27 Feb 2024

Equivariant polynomial ideals

This paper studies the existence and computation of finite bases for equivariant ideals of polynomials, which are invariant under renaming variables. Two main results are proven: a generalization of Hilbert's Basis Theorem showing these ideals are finitely generated, and an algorithm to compute a basis. Applications in computer science theory are discussed, like solving systems of equations and analyzing weighted automata.

Published on 27 Feb 2024

Classification of Positive Solutions for Sobolev Trace Inequality

This paper proves several classification results for the Euler-Lagrange equation of the Sobolev trace inequality, which determines the optimal constants. Using asymptotic estimates and integral identities, rather than symmetry arguments, the authors classify positive solutions under finite energy when $1 < p < n$.

Published on 27 Feb 2024

Detecting sleep from wrist motion data without ground truth labels

This paper introduces a new weakly supervised learning approach to detect sleep from wrist motion data, without requiring accurate sleep stage labels during model training. It uses an ensemble of simple existing sleep detection algorithms to generate 'weak labels', then trains neural networks on these labels. The method shows performance matching supervised models trained on ground truth labels, successfully enabling deep learning for sleep analysis without label data.

Published on 27 Feb 2024

Sustained emission from suspended WSe2 at low carrier densities

This paper investigates suspending monolayer WSe2 to reduce defects and sustain strong photoluminescence emission at low pumping powers. Suspending was found to substantially enhance photoluminescence intensity and quantum yield compared to contacted WSe2, especially exponentially more so at low carrier densities. This highlights the potential of using suspended WSe2 for developing efficient, low-power quantum light emitters.

Published on 27 Feb 2024

Localizing gravity via optimal shielding

This paper introduces a localized seed-to-solution method to construct solutions to Einstein's equations that exhibit gravity shielding, or localization. The solutions achieve optimal shielding with arbitrarily narrow gluing domains and arbitrarily slow metric/curvature decay, yet enjoy super-harmonic control. New concepts include normalized asymptotic kernel elements, localized energy functionals, localized ADM modulators, and relative energy-momentum vectors.

Published on 27 Feb 2024

Duality and identities for nonabelian group error-correcting codes

This paper proposes a definition of duality for nonabelian group codes based on the irreducible representations of the group. The authors prove analogues of Greene's Theorem and the MacWilliams Identities relating properties of a code to those of the proposed dual object. When specialized to abelian groups, their results match existing formulas in coding theory.

Published on 27 Feb 2024

Corridor model predictive control enables safe multi-agent inspection

This paper proposes an extension of Corridor Model Predictive Control (CMPC) to enable safe multi-agent inspection missions. CMPC leverages high order control barrier functions to maintain each inspector spacecraft within a predefined safe corridor relative to its assigned trajectory. Authors demonstrate recursive feasibility of the control scheme, introduce modifications to ensure safety in continuous time, and validate performance in a simulated inspection of the International Space Station involving trajectory tracking and collision avoid...

Published on 27 Feb 2024

Implicit regularization in neural networks for matrix sensing

This paper studies implicit regularization, which refers to neural networks' ability to find good solutions without explicit regularization terms in the loss function, in the context of non-linear neural networks for matrix sensing problems. The authors propose a Spectral Neural Network (SNN) architecture that represents matrices in terms of their singular values and vectors rather than entries. They prove SNNs demonstrate an implicit regularization phenomenon, ensuring gradient descent converges exponentially fast to the best approximation ...

Published on 27 Feb 2024

Simplified potential mapping reveals degeneracies in strong gravitational lensing

This paper explores degeneracies in interpreting gravitational lens models that can fit observations equally well. By thinking in terms of the projected potential rather than mass, equivalent models preserving properties at image locations are created via quadratic programming. This highlights ill-constrained regions and redistributes mass, but doesn't fully remove detected dark clumps. Generalized mass-sheet and source-plane transformations are also illustrated.

Published on 27 Feb 2024

The outflows, bubbles, and knots of NGC4418

New observations reveal several previously unknown features in the galaxy NGC4418, including outflows, bubbles, and ionized gas knots. These indicate that its extremely obscured nucleus is likely powered by an active galactic nucleus.

Published on 27 Feb 2024

Combating noisy labels with relaxed contrastive learning

This paper proposes an end-to-end framework called PLReMix to address issues with using contrastive representation learning (CRL) techniques in learning with noisy labels (LNL). It introduces a pseudo-label relaxed (PLR) contrastive loss that constructs reliable negative pairs for each sample, avoiding optimization conflicts. It also expands sample selection to a 2D Gaussian mixture model leveraging both semantic information and model outputs. Experiments show the method is effective and the PLR loss is scalable to boost other LNL techniques.

Published on 27 Feb 2024

Overview of CI/CD usage and evolution

This paper analyzes over 600,000 GitHub repositories to understand developers' usage of CI/CD technologies. It finds 32.7% of repositories use CI/CD, with GitHub Actions dominating at 57.8%. Newer languages like Go and Rust use CI/CD more than older ones like C and Java. 15.1% of projects use multiple CI/CD technologies simultaneously, some using up to 13. Technologies change frequently, with 7-12% of snapshots having CI/CD changes annually. Popular transitions are from Travis CI to GitHub Actions.

Published on 26 Feb 2024

Search for invisible particles in boosted single top quark events in proton-proton collisions

The paper presents a search for dark matter particles or heavy vector-like quarks produced in association with a single top quark. No significant excess over background was found. New exclusion limits were placed on the mass of a dark matter mediator particle up to 4.3 TeV and mass of a vector-like quark up to 1.8 TeV.

Published on 26 Feb 2024

Simplified point collocation method using smooth basis functions

This paper introduces a method for numerically solving partial differential equations using smooth basis functions constructed by convolving piecewise polynomials with a smoothing kernel. The smoothness and approximation properties of these basis functions make them well-suited for solving PDEs through point collocation, where the PDE is directly evaluated at a set of points. This avoids the need for domain and boundary integration required in Galerkin finite element methods. The paper shows convergence of the method when applied to Poisson,...

Published on 26 Feb 2024

Tests of Macrorealism in Discrete and Continuous Variable Systems

I study several aspects of tests of macrorealism (MR), examining the necessary and sufficient conditions, generalizing to arbitrary measurement times and many-valued variables, analyzing conditions of MR, developing tests in continuous variable systems through position coarse-grainings, calculating temporal correlators and LG violations in the quantum harmonic oscillator, and outlining a different approach to meeting the invasiveness requirement of LG tests.

Published on 26 Feb 2024

Oscillating states of Brownian motion under weak periodic drives

This paper proposes a mathematical method to analyze the behavior of a Brownian particle subjected to periodic but weak drives, where the viscosity and noise approach zero. It finds two classes of distributions for the particle's position and velocity. One class tends to equilibrium, with tunable temperature useful for applications like cooling. The other class resonates at fractional periods, which could trigger chemical reactions.

Published on 26 Feb 2024

Magnetic liquid channels for low-friction flow

This paper demonstrates the creation of microfluidic channels using magnetic fields to confine ferrofluids, surrounding a transported liquid. These liquid-liquid channels of under 10 μm diameter enable very low friction flow. Tracking experiments directly confirm the high wall velocities that enable reduced shear.

Published on 26 Feb 2024

Cross-lingual open retrieval QA

This paper proposes a self-supervised method to train a single encoder-decoder model for cross-lingual question answering. It uses the cross-lingual links in Wikipedia to create training data, avoiding the need for annotated datasets or machine translation. The model, CLASS, shows state-of-the-art performance in supervised and zero-shot settings.

Published on 26 Feb 2024

Determining parameters of Barrow dark energy model

This paper proposes using a cosmographic approach to determine key parameters of the Barrow entropic dark energy cosmological model. Rather than a complex search through parameter space, the model parameters are directly calculated from observable characteristics of the universe's expansion. The Barrow model aims to account for quantum gravitational effects on the horizon entropy and dark energy density. By relating the parameters to current measurements of expansion dynamics, represented by cosmographic parameters, this approach avoids exte...

Published on 26 Feb 2024

Semantic image generation from noisy labels

This paper proposes a robust conditional diffusion model called Stochastic Conditional Diffusion Model (SCDM) for generating photo-realistic images from semantic label maps. It features a novel forward process that stochastically perturbs the input labels through a technique called Label Diffusion, making noisy labels become more similar to clean labels over time. This allows the model to generate images close to what would be produced from clean labels. A class-wise noise schedule is also proposed to handle small, rare objects better. Exper...

Published on 26 Feb 2024

LLM memory performance

This paper investigates how well large language models perform on memory tests designed for humans. It focuses specifically on recognition and recall of words from lists, comparing LLMs to past results from human subjects.

Published on 26 Feb 2024

Gluon structure of the proton

This paper uses a spectator model to examine gluon distributions inside the proton. It studies gluon generalized parton distributions at both zero and non-zero skewness, representing gluons in 3D momentum space. The model also enables calculating gluon contributions to proton spin and orbital angular momentum.

Published on 26 Feb 2024

Shell model study of beta decay in Barium isotopes

This paper performs a systematic theoretical study of the beta decay properties in Barium-139, 140, and 141 isotopes using the nuclear shell model. Calculations match well with recent experimental data on Barium-141 decay. Key results include theoretical interpretation of measured beta decay log(ft) values, half-lives, and average shape factors.

Published on 26 Feb 2024

Transformer Network for Autonomous Vehicle Trajectory Prediction

This paper introduces a transformer network model to predict surrounding vehicles' future trajectories for autonomous driving. It incorporates map context using a CNN and penalizes infeasible off-road predictions. Experiments show state-of-the-art accuracy and feasibility.

Published on 26 Feb 2024

Controlling quantum anomalous Hall band gap via magnetization rotation in TiSb monolayer

Researchers establish guidelines between symmetries and orbital properties that allow tuning the band gap in quantum anomalous Hall materials. Following this framework, they computationally predict a TiSb monolayer exhibits a band gap adjustable from 0 to 209 meV based on magnetization orientation.

Published on 26 Feb 2024

Follow-up of ANTARES neutrino alerts reveals constraints on transient sources

The ANTARES neutrino telescope has operated an alert system since 2009 to follow up neutrino event candidates with space- and ground-based telescopes across the electromagnetic spectrum. Results from 12 years of follow-up campaigns are presented. No significant counterparts were found, allowing constraints to be placed on neutrino emission from transient sources like gamma-ray bursts and core-collapse supernovae. A fortuitous association between one neutrino alert and an X-ray flare was identified and studied. This program and its lessons le...

Published on 26 Feb 2024

Decoupling sanitization from fuzzing

This paper presents SAND, a new fuzzing framework that separates expensive sanitization checks from the main fuzzing loop. SAND performs fuzzing on an unmodified program binary and only invokes sanitizer instrumentation when an input exhibits a unique execution pattern not seen before. This allows most fuzzing time to be spent on fast unmodified code. SAND uses an efficient 'execution pattern' to approximate execution paths and identify unique inputs needing sanitization. Evaluated on real-world programs, SAND achieves 2.6x and 15x higher th...

Published on 26 Feb 2024

Representations and cohomology of matched pairs

This paper introduces representations and cohomology theories for matched pairs of Lie algebras, which are structures that arise in the study of Lie bialgebras and Poisson-Lie groups. The cohomology links to deformations and extensions. Matched pairs are also generalized to L-infinity algebras, with connections to cohomology.