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.
Published on 8 May 2024
Radiative corrections modify tracer particle dispersion
We study a tracer particle coupled to a scalar Bose field in the limit of infinite field propagation speed. For initial states without field excitations, we derive an effective dynamics where the particle propagates with a modified dispersion relation. This can be understood as radiative corrections due to interactions with virtual bosons.
Published on 8 May 2024
Atmospheric turbulence mitigation in wide-field images
This paper introduces a new technique to remove the effects of atmospheric turbulence from wide-field astronomical images. It uses a neural network trained on simulated data that takes in a sequence of short-exposure images of a star field and reconstructs a single sharp, noiseless image. Across different levels of atmospheric seeing, it accurately associates speckles with their source stars and separates light from neighboring stars. This gives much clearer images without compromising astrometric stability or flux measurements.
Published on 8 May 2024
Particle dynamics with heterogeneous rates
This paper studies a model of particles moving on a line, where each particle jumps randomly left and right at its own intrinsic rates. An exclusion interaction prevents multiple particles occupying one site. Under conditions on the jump rates, if started closely packed, the particles form a 'stable cloud', meaning the gaps between particles reach a stationary distribution, and all particles share one overall speed.
Published on 8 May 2024
Predicting Cellular Traffic in Real Time
This paper investigates methods for accurately forecasting cellular network traffic volumes in real-time scenarios, examining two live prediction algorithms applied to machine learning models. The study reveals that the Fast LiveStream Prediction (FLSP) algorithm enhances bandwidth efficiency, accuracy, and processing load compared to traditional rolling algorithms when data is reported asynchronously across the network. Through analysis and simulation, the paper provides guidance on model selection and data gathering strategies to optimize ...
Published on 8 May 2024
Monte Carlo confidence intervals
This paper presents an efficient method to construct exact or conservative confidence intervals from Monte Carlo hypothesis tests. It uses a single set of simulations to test all parameter values, reducing computation. For real-valued parameters, intervals can be found quickly when the p-value is quasiconcave. Additional savings are possible for common statistics in one-sample and two-sample problems. An open-source Python implementation is provided.
Published on 8 May 2024
Predicting Bursty Traffic in M2M Networks
This paper presents a machine learning framework to forecast bursty traffic patterns in massive machine-type communication networks. It uses long short-term memory and dense neural networks to accurately predict traffic and identify congestion events by assimilating real-time data. A new online prediction algorithm updates model states efficiently. Evaluations demonstrate 52% higher accuracy without additional overhead.
Published on 8 May 2024
Classical angular momentum in multi-body systems
This paper analyzes the concept of grand angular momentum, used in quantum studies of systems with multiple bodies, and applies it to classical physics problems. Using tree diagrams to represent different coordinate systems, the authors decompose the grand angular momentum into regular 3D angular momenta. This allows generalizing some 2-body results, and deriving an expression for the scattering angle in a general multi-body system.
Published on 8 May 2024
Airborne transmission between people at tables
This study uses simulations to explore how airborne transmission of viruses like COVID-19 can occur when people are seated facing each other at a table, like in restaurants or meetings. It finds that the presence of the table changes exhaled flows during breathing, speech and laughter in ways that can either restrict forward spread of pathogens or increase their concentration for a facing person, raising transmission risk. The table also filters out medium-sized droplets, redirecting them to surface contact transmission.
Published on 8 May 2024
Efficient GNN training on disk
This paper introduces DiskGNN, a system to efficiently train graph neural networks on disk when graphs exceed CPU memory. DiskGNN achieves high efficiency and model accuracy through offline sampling to optimize data layout, four-level caching, batched packing, and pipelined training. Experiments show DiskGNN speeds up state-of-the-art systems by over 8x while matching accuracy.
Published on 8 May 2024
Efficiency of basic and applied research across European countries
This paper analyzes the efficiency of 28 European Union countries in conducting basic and applied research from 2008-2014. It uses a robust data envelopment analysis method to classify countries as efficient or inefficient, finding that more economically advanced countries tend to have higher research efficiency, especially for applied research with more tangible outputs.
Published on 8 May 2024
Language-guided robot control for surgical tasks
This paper presents SuFIA, a framework that uses large language models (LLMs) and perception modules to plan and execute robotic control for surgical sub-tasks. This allows for a learning-free approach to surgical automation without needing motion primitives or examples. SuFIA incorporates re-planning and human oversight to mitigate errors. Experiments in simulation and on a physical robot platform demonstrate SuFIA's ability to autonomously perform common surgical tasks under challenging conditions.
Published on 8 May 2024
Online Platform Content Moderation Policy Study
This paper analyzes content moderation policies from 43 major online platforms to understand their approaches to moderating copyright infringement, harmful speech, and misleading content. Using a custom web scraper and unified annotation scheme, the authors find significant variation across platforms and topics attributable to differing legal regimes. The paper lays groundwork for studying evolving moderation policies and their impacts.
Published on 8 May 2024
Accelerating diffusion models with distillation for fast high-quality image generation
This paper proposes a distillation framework to accelerate diffusion models, enabling high-quality and diverse image generation using only 1-3 sampling steps. Key innovations include Backward Distillation to reduce train-test discrepancy, Shifted Reconstruction Loss to transfer both structure and detail knowledge, and Noise Correction to enhance initial sample quality.
Published on 8 May 2024
Failure of the Harnack inequality for nonlocal kinetic equations
The authors prove that the Harnack inequality, an important regularity estimate, fails for nonlocal kinetic equations that arise as models for the Boltzmann equation without cutoff. This is in contrast to local kinetic equations that do satisfy the Harnack inequality. A counterexample is provided for the fractional Kolmogorov equation.
Published on 8 May 2024
XMM-Newton observations of the jet power and particle acceleration in the microquasar S26
The microquasar S26 has very powerful jets, indicating a super-Eddington accretion rate. However, X-ray observations show much lower luminosity than expected. We analyze XMM-Newton data, propose the disk emission is absorbed, additional spin energy powers the jet, and show particles can reach PeV energies.
Published on 8 May 2024
Duration analysis of unemployment benefit policy changes
This paper proposes methods to adapt difference-in-differences analysis to settings with duration outcomes, which indicate whether an individual has entered an absorbing state. It retains the simplicity of diff-in-diff while avoiding bias. The approach is applied to study an Austrian policy extending unemployment benefits, using a transparent cross-cohort design.
Published on 8 May 2024
Efficient attention computation for transformers
This paper develops a convolution-based method to efficiently approximate attention in transformers, reducing the quadratic complexity to nearly linear. It shows any attention matrix can be decomposed into convolution matrices, which enables fast Fourier transform for faster computation without changing model parameters.
Published on 8 May 2024
Clustering Retail Products by Customer Behavior
This paper proposes a new method to categorize retail products based solely on customer purchase data. It formulates product clustering as an optimization problem, using a genetic algorithm on market basket data to assign products to a given number of clusters. Key assumptions are that customers generally purchase one product per category, and products frequently purchased together should not be clustered together. Tests on simulated and real drugstore data from Czechia demonstrate the method can identify categories similar to those defined ...
Published on 8 May 2024
Text-driven 3D human pose estimation
This paper proposes FinePOSE, a new diffusion model-based approach for estimating 3D human poses from 2D keypoints. It introduces a novel fine-grained part-aware prompt learning mechanism to provide precise guidance for each human body part's movement. FinePOSE also establishes communications between the learned prompts and poses to enhance the diffusion model's denoising capability. Experiments show state-of-the-art performance on public benchmarks. An extension to multi-human scenarios also demonstrates potential.
Published on 8 May 2024
Benchmarking quantum gates using fixed-length sequences
The paper introduces a new randomized benchmarking protocol that uses fixed-length sequences of directly accessible gates, instead of relying on complex, compiler-optimized operations. This allows for fairer assessment of hardware performance. Key results quantify overestimation of fidelity by standard techniques. The method could enable building better noise models to advance error mitigation.
Published on 8 May 2024
SPIDER: Fast rank and select queries
This paper introduces SPIDER, a new succinct data structure for answering rank and select queries on bit vectors. SPIDER uses only 3.82% extra space, yet outperforms prior methods. For rank queries, SPIDER is the fastest known method on large inputs. For select queries, it narrows the performance gap between space-efficient and less space-efficient techniques. Key ideas include interleaving metadata with the bit vector to improve cache performance, and using predictions to accelerate select queries.
Published on 8 May 2024
Fast simulations for elastic solid dynamics
This paper proposes an efficient computational method to simulate the dynamics of elastic solids over time. It reformulates the equations that govern elastic solid behavior into a coupled system of equations. This allows the use of an exponential propagator technique, which provides faster and more stable solutions compared to traditional time-stepping methods. The new approach is demonstrated to enable much larger time-steps while maintaining accuracy, leading to dramatic speedups in computation time.
Published on 8 May 2024
Modeling single-pion production in neutrino interactions
This paper presents a new model for single-pion production in neutrino interactions implemented in the NuWro Monte Carlo generator. The model combines contributions from nucleon resonances and non-resonant backgrounds, providing improved predictions compared to previous versions across the full kinematic range up to deep inelastic scattering. When incorporated into NuWro's nuclear framework, accounting for initial and final state effects, the new model shows better agreement with recent accelerator neutrino experiment data on pion production.