Published on 14 May 2024
Quantum oscillations reveal electron pocket in underdoped cuprates
Researchers propose a model where pseudogap 'Fermi arcs' in cuprates reconstruct into a single electron pocket with charge density wave order in high magnetic fields. Computations show this matches quantum oscillations seen in experiments.
Published on 14 May 2024
Advancing driving perception technologies under challenging real-world conditions
The 2024 RoboDrive Challenge focused on innovating robust perception systems for autonomous vehicles that can withstand diverse disturbances like weather changes and sensor failures. 140 international teams participated, pushing boundaries. Key innovations emerged in data augmentation, sensor fusion, self-supervised learning, and new algorithms that handle sensor inconsistencies and environment variability more effectively. Extensive analysis of solutions provided insights to guide future research towards safer, more reliable autonomous syst...
Published on 14 May 2024
Efficient vision-language learning with cluster masking
This paper proposes a simple yet effective strategy for masking image patches during visual-language contrastive learning. By randomly masking clusters of visually similar patches in each training iteration, the model is forced to predict words for missing visual structures using context. This provides an extra learning signal and speeds up training. When evaluated on several benchmarks, the proposed approach outperforms strategies like random patching (FLIP) in representation quality and downstream performance.
Published on 14 May 2024
Performance of wave function and Green's function methods for driven non-equilibrium dynamics
This paper compares different theoretical methods for simulating non-equilibrium quantum dynamics, driven by an intense electric field pulse. It focuses on wavefunction-based methods like configuration interaction and coupled cluster theory, and Green's function methods using approximations like GW. Under strong driving, systems can exhibit strongly correlated behavior not captured well by methods designed for weak correlations. The coupled cluster method performs excellently for weak and moderate perturbations but struggles for strong drivi...
Published on 14 May 2024
Long video comprehension benchmark
This paper introduces CinePile, a novel benchmark for evaluating long-form video understanding models. It contains over 300,000 multiple choice questions covering various aspects of visual, temporal, and multimodal comprehension. Models are tested on their ability to reason about events, human-object interactions, and plot progressions in long video scenes based on both visual and dialogue information. Even state-of-the-art models lag significantly behind human performance, highlighting remaining challenges in long-form video understanding.
Published on 14 May 2024
Chaotic dynamics in the basin of attraction boundary via non-transversal intersections
This paper analytically proves the existence of transversal homoclinic points and a chaotic invariant Cantor set on the boundary of the basin of attraction for a non-globally smooth diffeomorphism map related to the Secant method. The map lacks key properties like smooth invertibility, but the authors adapt tools like the Smale-Birkhoff theorem to show conjugacy with a full shift map on the chaotic set.
Published on 14 May 2024
Qiskit software for quantum computing
Qiskit is an open-source software development kit for quantum computing research, education, and applications. It provides tools for composing, simulating, optimizing and running quantum circuits, targeting various hardware backends. Key features highlighted include modular design, performance considerations, hardware portability, multiple abstraction levels, and integration of quantum and classical computing.
Published on 14 May 2024
Cancelling anomalies in 8D gauge theories from 7-branes
This paper studies an 8D N=1 gauge theory with gauge group Sp(k) that arises from k D7-branes probing an O7+-plane. For k>1 this gauge theory has a subtle anomaly that the authors conjecture can be cancelled by coupling to an 8D topological field theory (TFT). In this work, the required TFT is explicitly constructed by placing D3-branes and fluxbranes infinitely far from the 7-branes. The topological operators from these branes have vector bundles defined on them, and the KO/KSp-homology classes of these bundles play a role in the anomaly ca...
Published on 14 May 2024
Localized energy density of gravitational waves
This paper proposes a new set of charges at null infinity that allow defining a localized energy density of gravitational waves. This is achieved by improving the corner of the symplectic structure to obtain charges with a 'hard' flux. They satisfy standard criteria like covariance and vanish in non-radiative spacetimes. The localized energy density provides an operational way to measure the energy absorbed locally from a gravitational wave flux.
Published on 14 May 2024
Sum of Random Variables
This paper revisits a classic probability theory problem posed by Kolmogorov on characterizing the possible distributions for the sum of two random variables given their marginal distributions. The authors clarify and correct confusion in prior literature on the sharpness of bounds derived by Makarov and Frank et al. They also connect this theory to bounding distributions of individual treatment effects, sharpening some previous results of Fan & Park.
Published on 14 May 2024
Temperature effects on structure of Ruddlesden-Popper nickelate La3Ni2O7
This paper examines how temperature impacts the crystal structure of the Ruddlesden-Popper bilayer nickelate La3Ni2O7 (LNO-2222) using laboratory X-ray diffraction on single crystals. Contrary to prior work suggesting increased symmetry at lower temperatures, enhanced tilt distortions are observed, indicating potential lowering of symmetry and favorability of the Cmcm space group. This establishes a benchmark methodology and provides insights into the complex structural dynamics of this system.
Published on 14 May 2024
Variational Bayes as a model of belief updating
This paper shows how variational Bayesian methods can provide a foundation for modeling how people update beliefs in a non-Bayesian way. By modifying the variational Bayes objective function, the author derives a popular exponential belief updating rule used in behavioral economics. The paper explores attitudes toward uncertainty and model misspecification that map to parameters of this exponential rule.
Published on 14 May 2024
Locally interacting stochastic differential equations driven by fractional Brownian motion
We consider collections of stochastic differential equations indexed by a graph, with each equation driven by fractional Brownian motion. The drift term in each SDE interacts only with the SDEs corresponding to neighboring vertices. We derive a fundamental martingale for the driving noise which allows a Girsanov-type change of measure. This leads to weak existence, uniqueness, and a 2-Markov Random Field property for the laws of these interacting systems.
Published on 14 May 2024
A Simplified Introduction to Causal Inference in Machine Learning
This lecture note provides a simplified introduction to causal inference for machine learning students without prior exposure. It focuses on expanding their view of machine learning to incorporate causal reasoning for better out-of-distribution generalization.
Published on 14 May 2024
Simplifying Aviation Rules with AI
This paper develops the first database of expertly-labeled question-answer pairs on Colombian aviation regulations and uses it to train AI models to simplify those complex rules. By translating technical jargon into plain language, the AI aims to improve understanding and compliance in the aviation industry.
Published on 14 May 2024
Critical three-cycle dynamics for secant map model
This paper analyzes a model map encoding the dynamics of the secant root-finding method near critical periodic points. The model's basin of attraction shape depends on parameter parity and sign. Boundaries may contain stable manifolds of fixed points or two-cycles.
Published on 14 May 2024
Kolmogorov-Arnold Networks for Satellite Traffic Forecasting
This paper proposes using Kolmogorov-Arnold Networks (KANs), a novel neural network architecture, for satellite traffic forecasting. KANs leverage spline-based activation functions that can learn complex patterns, replacing traditional linear weights. When tested on real-world satellite data, KANs significantly outperformed conventional Multi-Layer Perceptrons (MLPs), providing more accurate forecasts using far fewer trainable parameters. An ablation study also explores how KAN-specific parameters impact performance. Overall, this demonstrat...
Published on 14 May 2024
Ranking graph transformations to repair inconsistencies
This paper presents an approach to use graph transformation rules to repair inconsistencies in graphs. It allows for inconsistencies to exist for a while before repairing them when needed. The key idea is to equip rules with impairment- and repair-indicating application conditions that count violations of constraints, instead of blocking rule applications. A main result shows the difference in constraint violations before and after a rewrite step equals the difference between violations of impairment- and repair-indicating conditions. This e...
Published on 14 May 2024
Incorporating Clinical Rules into AI for Prostate Cancer Diagnosis
This paper proposes a new approach to integrate clinical guidelines into AI models for prostate cancer diagnosis, without needing extra annotations or model parameters. It adapts a large multi-modal language model using a two-stage process: first to handle MRI inputs, then to incorporate the guidelines as instructions. By distilling these instruction-guided features into the diagnosis model, it aligns the model with expert knowledge in the guidelines. Evaluated on real clinical data, this approach improved model accuracy and clinical applica...
Published on 14 May 2024
Absorbed central engine revealed in Mrk 1239
Deep X-ray observations of the NLS1 galaxy Mrk 1239 reveal complete absorption of the AGN continuum below 3 keV by a 10^23.5 cm^-2 neutral medium. Timing analysis shows no variability in soft X-rays, indicating a large-scale origin. A strong flare seen in hard X-rays is consistent with blurred reflection off the accretion disc. Collisionally ionized and photoionized emission may originate from an outflow crashing into surrounding material.
Published on 14 May 2024
Refining an epilepsy dictionary for social media
Researchers built an epilepsy dictionary from medical sources to analyze Instagram posts. They had human annotators evaluate dictionary term matches and found 8 frequent but ambiguous terms. After removing those terms, network analysis showed improved identification of medically relevant terms in the social media cohort.
Published on 14 May 2024
Fast deep learning pipeline for neonatal cortical surface extraction
This paper proposes a fast deep learning pipeline to extract cortical surfaces from neonatal brain MRIs in the Developing Human Connectome Project dataset. A multiscale deformation network is introduced to predict cortical surfaces end-to-end without requiring tissue segmentation. This pipeline incorporates GPU-accelerated processing and completes within 24 seconds, nearly 1000 times faster than the original 6.5 hour pipeline. Manual evaluation shows this pipeline produces superior or equal cortical surface quality for 82.5% of test cases.
Published on 14 May 2024
Searches for exotic particles at CERN
This paper summarizes searches by the ATLAS and CMS experiments at CERN's Large Hadron Collider for three types of hypothetical exotic particles that go beyond the Standard Model of particle physics: heavy neutral leptons, long-lived particles, and vector-like quarks. No signals have been observed yet, but improved limits have been set, constraining some theoretical models and mass/lifetime ranges.
Published on 14 May 2024
Total coloring via efficient dominating sets
This paper studies total colorings of graphs where each vertex color class forms an efficient dominating set, meaning a minimal set that is within 1 edge of every vertex. It proves the 3-cube graph has such a coloring using 4 colors, and conjectures this is the only such case. It relates this to edge colorings of prism graphs formed from star transposition graphs.
Published on 14 May 2024
Forecasting eye disease progression from longitudinal images
This study proposes a deep learning model to predict patients' future risk of developing vision-threatening eye diseases like age-related macular degeneration and glaucoma. By analyzing sequences of retinal images captured over time, the model learns to assess disease progression rates and make personalized risk forecasts to inform care plans.
Published on 14 May 2024
Learning neural Granger causality with a single model
This paper proposes a new method to identify causal relationships between time series variables using a single neural network model. By regularizing the input-output Jacobian matrix, the method eliminates issues with relying on weight sparsity and separate models per variable. It demonstrates high performance and scalability in capturing both multivariate and temporal causal effects on benchmark datasets.
Published on 14 May 2024
Quantum systems from separation of variables on sphere
This paper studies quantum integrable systems arising from separation of variables in orthogonal coordinates on a 3-sphere. The analysis focuses on the resulting differential equations and their polynomial solutions. A key finding is that separation in prolate coordinates leads to a system with quantum monodromy, indicating a global obstruction to assigning quantum numbers.
Published on 14 May 2024
Daydreaming neural networks effectively store patterns
Researchers developed an iterative neural network training procedure called Daydreaming that continuously reinforces memories to store while removing spurious memories. On both synthetic random data and real-world MNIST images, Daydreaming networks matched or exceeded state-of-the-art storage capacity and retrieval quality, even exploiting correlations to improve performance. The networks developed high-quality attractors matching unseen examples and class averages.