Published on 9 May 2024
Departures from ideal black hole gravitational wave scattering
This paper proposes a way to parameterize hypothetical modifications to gravitational wave scattering by black holes, allowing model-independent tests. It shows how such modifications would alter gravitational wave signals from extreme mass ratio inspirals. Even small changes to scattering can accumulate over orbits, enhancing sensitivity. Initial estimates suggest future detectors could probe extremely tiny deviations.
Published on 9 May 2024
Learned Bayesian evidence estimation with normalizing flows
This paper presents a method to estimate the Bayesian evidence, a quantity used in model comparison, using normalizing flows. The method trains a normalizing flow on posterior samples to learn an internal importance sampling target distribution. This solves issues with the variance exploding in the original harmonic mean estimator. The resulting estimator is more robust, flexible and scalable than previous approaches.
Published on 9 May 2024
Analysis of growth rates for surrogate loss bounds in classification
This paper analyzes the growth rates of bounds relating the estimation error of surrogate losses to that of the 0-1 loss, known as H-consistency bounds, in both binary and multi-class classification. It shows these bounds exhibit a universal square-root rate near 0 for common smooth losses like cross-entropy, under mild assumptions. This implies directly reducing the surrogate estimation error leads to a square-root decay in the target 0-1 estimation error. The paper also thoroughly compares losses based on minimizability gaps, which become ...
Published on 9 May 2024
Accelerating diffusion models for fast image generation
This paper proposes a method to distill a complex, multi-step diffusion model into a fast, single-step conditional GAN model that can generate images nearly as well but much more quickly. Their key ideas are to interpret diffusion distillation as an image-to-image translation task using noise and image pairs from the diffusion model, and to create losses that work directly in the model's latent space, avoiding decoding to pixels. Their one-step model outperforms other state-of-the-art diffusion distillation techniques.
Published on 9 May 2024
Slow rotation of long bars in galaxies concentrates star formation along them
This paper studied how the strength and rotation speed of bars in disc galaxies affect star formation in different regions. It found that long, slowly rotating bars concentrate gas and star formation along themselves. Short or fast bars do not have the same effect. The results suggest bars influence galaxies most when they are both long and slowly rotating.
Published on 9 May 2024
Evaluating Language Models for Driving
This paper provides a comprehensive analysis of state-of-the-art multimodal language models in simulated driving environments. The models demonstrate significant limitations in reasoning logically about basic vehicle dynamics, interactions between vehicles, trajectory planning, and unexpected events across sequences of images. To enable analysis, the authors introduce DriveSim, a specialized driving simulator generating diverse scenarios. They also release full code and a dataset for continued research. Results reveal critical gaps in curren...
Published on 9 May 2024
Transforming Text into Images, Videos, 3D Objects and Audio via Flow-based Diffusion Transformers
This paper introduces Lumina-T2X, a family of flow-based large diffusion transformers designed to transform noise into images, videos, 3D objects and audio conditioned on text. Key techniques like tokenized representations, learnable placeholders, RoPE, RMSNorm and flow matching enable unified training and flexible generation across modalities and resolutions. Models scale up to 7B parameters with 35% of the training costs of a 600M model, achieving ultra-high-res images and long 720p videos.
Published on 9 May 2024
Interlayer twisted graphene forms flat bands
Researchers created multilayer graphene structures by twisting a monolayer graphene sheet onto a multilayer Bernal-stacked graphene flake. They showed that applying an electric field isolates a flat, topological band localized primarily in the graphene layers near the twisted interface. This produced similar correlated electronic phases across twisted structures, enabling new opportunities to study exotic phenomena by tuning the flat band with layer number.
Published on 9 May 2024
Forbidden induced subgraphs in sparse graphs
We study the maximum number of edges possible in a subgraph of a sparse graph that avoids containing a fixed bipartite graph as an induced subgraph. We determine tight bounds when the sparse graph is a random graph or Paley graph and the forbidden subgraph has bounded degree. As an application, we resolve a question of Alon, Krivelevich and Samotij by showing that hereditary properties missing a bipartite graph have subgraphs of quadratically fewer edges in random graphs.
Published on 9 May 2024
Construction of a family of approximate formulations for the stationary Stokes problem using an extended system
This paper introduces an exact parameterized extended system that contains the solution of the weak formulation of the homogeneous Dirichlet problem for the stationary Stokes equations between its solution components. It reformulates the consistent pressure Poisson equation from the unsteady to the stationary case to retain information for the approximate pressure on the boundary. A parameterized perturbed pressure Poisson equation is obtained. It is proven equivalent to solve the Stokes problem or to solve a problem with the momentum equati...
Published on 9 May 2024
Computation of spiral wave spectra using exponential weights
This paper shows that exponential weights serve as effective preconditioners to bound the resolvent and enable accurate computation of spiral wave spectra. The optimal exponential rates can be determined from a simpler 1D asymptotic problem.
Published on 9 May 2024
Volume entropy bounds for harmonic manifolds
This paper examines a class of non-compact harmonic manifolds called harmonic manifolds of hypergeometric type, which includes spaces like rank one symmetric spaces and Damek-Ricci spaces. By normalizing the metric to satisfy Ricci curvature = -(n-1), the author shows the volume entropy is bounded between 2√(2/3)(n-1) and (n-1). The upper bound is achieved only by real hyperbolic spaces, while the lower bound is achieved in 4 cases of Damek-Ricci spaces.
Published on 9 May 2024
Efficient ranking of text options through selective pairwise comparisons
This paper proposes a framework to efficiently rank a set of text options by quality. It uses selective pairwise comparisons judged by a language model, combining these decisions probabilistically. With just a small subset of all possible comparisons, it can predict quality scores that correlate well with human judgment, while greatly reducing computational costs.
Published on 9 May 2024
Role of vanadium oxide layer in superconducting properties of strontium vanadium iron arsenide
This paper investigates how oxygen deficiency in the vanadium oxide layers of the iron-based superconductor Sr2VFeAsO3-δ affects its electronic properties. Increased oxygen deficiency elongates the crystal c-axis and suppresses superconductivity. A distinct magnetic and structural anomaly appears around 100 K that indicates orbital ordering of vanadium dxz/dyz states. Fluctuations related to this ordering seem to propagate to the FeAs layers and impact superconductivity.
Published on 9 May 2024
Quantum architecture search for optimal circuit design
This paper proposes a reinforcement learning approach for automatically searching optimal quantum circuit architectures. It uses a recurrent neural network controller to sample circuit designs, adding layers sequentially. Performance feedback on a classification task trains the controller. This allows efficient exploration to find high-performing, low-complexity circuit architectures outperforming manual design.
Published on 9 May 2024
Financial knowledge reduces self-imposed borrowing constraints
This paper provides evidence that less financially knowledgeable entrepreneurs face higher costs and pessimism when seeking loans, making them more likely to discourage themselves from applying. Their tendency to avoid applying, despite needing financing, represents an inefficient self-rationing that financial skills could alleviate.
Published on 9 May 2024
Safe Reinforcement Learning Using Uncertainty-Aware Models
The paper proposes a new reinforcement learning method called CERL that maintains safety while learning policies, using Bayesian neural networks to model uncertainty and suggest policies robust to inaccuracies. CERL outperformed other methods on constrained MDP tasks from image inputs.
Published on 9 May 2024
Finite element methods for simulating evolution of axisymmetric solid thin films
Researchers developed mathematical models and numerical methods to simulate the process of solid thin films evolving into isolated islands over time. This occurs by a process called solid-state dewetting. They introduced new numerical techniques that are applicable to more surface energy scenarios, provably preserve key physical properties, and generate quality mesh. Numerous tests demonstrated the accuracy and efficiency.
Published on 9 May 2024
Simultaneously Reducing Tensions in Hubble Constant and Matter Density
This paper introduces a K-essence cosmological model with a coupling between the scalar field potential and kinetic terms. It shows this model can simultaneously reduce tensions between early and late Universe measurements of the Hubble constant (to 2.2 sigma) and matter density fluctuations (to 0.82 sigma).
Published on 9 May 2024
Self-supervised modeling for text recognition
This paper proposes Symmetric Superimposition Modeling (SSM), a self-supervised approach for text recognition that captures both character shapes and linguistic context by reconstructing original and inverted images from their superimposition. SSM operates at both pixel and feature levels. At the pixel level, it reconstructs original and inverted images to capture shapes and texture-level context. At the feature level, it reconstructs features of the original and inverted images under different augmentations to model semantic-level context a...
Published on 9 May 2024
Emulator for forecasting matter clustering in f(R) gravity models
This paper introduces an emulator called FREmu to swiftly and accurately predict nonlinear matter power spectra in f(R) gravity models, which modify general relativity and impact structure formation. By training on simulation data, it maps a 7D parameter space to power spectra using principal component analysis and neural networks. It achieves 95%+ accuracy across scales and redshifts, enabling efficient cosmological parameter constraints.
Published on 9 May 2024
Pseudoscalar explains muon magnetic moment anomaly
This study examines whether a light pseudoscalar particle predicted by extensions of the Standard Model called Two Higgs Doublet Models (THDMs) can account for the anomalous magnetic moment of the muon. Considering experimental constraints from B-meson decays, meson mixing, and Higgs decays, the authors find a 66 GeV pseudoscalar with a high ratio (58) of two Higgs vacuum expectation values can resolve the muon g-2 anomaly.
Published on 9 May 2024
Phase-space particle rearrangement bounds
This paper discusses theories to calculate bounds on the effects of plasma phenomena that rearrange populations of particles in phase space. Two key problems are understanding the mapping between rearrangement rules and possible outcomes, and determining appropriate rules for different systems. Recent progress is summarized, but open questions remain.
Published on 8 May 2024
Event-based open-vocabulary scene parsing
This paper introduces OpenESS, a method to perform event-based semantic segmentation with open-ended textual queries instead of fixed labels. It transfers knowledge from image and text models to event data, allowing segmentation of new categories without retraining. Key techniques include contrastive learning between events and image regions, and optimizing event embeddings to match text meanings.
Published on 8 May 2024
Data-efficient 3D scene understanding for autonomous vehicles
This paper proposes a semi-supervised framework called LaserMix++ that leverages both LiDAR point clouds and camera images to improve 3D scene understanding for autonomous driving with far less labeled data. Key innovations include multi-modal data mixing, transferring knowledge from images to point clouds, and generating auxiliary labels from language models, which enhance regularization and feature learning.
Published on 8 May 2024
Evaluating and Reducing Hallucinations in Vision-Language Models
The paper proposes THRONE, a new benchmark to evaluate 'Type I' hallucinations (in open-ended responses) in large vision-language models (LVLMs). It utilizes language models to identify hallucinations and introduces metrics to quantify them. The paper demonstrates that reducing 'Type II' hallucinations (in responses to specific questions) does not reduce Type I hallucinations, and that existing methods for evaluating Type I hallucinations are limited. Finally, a simple data augmentation method is introduced that reduces both Type I and Type ...
Published on 8 May 2024
Using diffusion models for cosmological parameter inference
This paper trains a diffusion model to generate simulations of cosmic density fields based on input cosmological parameters. It then uses the model's likelihood estimate to infer those parameters from a given density field, showing this approach yields tight constraints and is robust to noise perturbations.
Published on 8 May 2024
Decoder-decoder architecture for efficient language models
The paper proposes YOCO, a decoder-decoder architecture for large language models. It consists of a self-decoder that encodes global key-value caches, and a cross-decoder that reuses those caches. This design reduces GPU memory usage and speeds up inference compared to regular Transformer decoders. Experiments show YOCO scales well in terms of model size, training data, and context length. At 1 million tokens it achieves high accuracy on retrieval tasks. Profiling shows orders of magnitude less memory usage and faster prefilling.