Clear Signals: Forecasting Lake Health with Sparse Data
A new machine learning framework pinpoints the minimal data needed to accurately predict lake water clarity, streamlining monitoring efforts and improving resource management.
Science
Leveling the Playing Field: Boosting Rare Classes with Confidence-Guided Learning
![The weighting function [latex]\Omega(p_t, f_c)[/latex] demonstrates an asymmetric relationship between prediction confidence [latex]p_t[/latex] and class frequency [latex]f_c[/latex], utilizing a pivotal point ω to differentiate low-confidence regions and facilitate smooth transitions between patterns characteristic of both tail and head classes.](https://arxiv.org/html/2601.15924v1/wireframe.png)
A new approach dynamically adjusts training to prioritize challenging examples from underrepresented classes, improving performance on imbalanced datasets.
Skyborne Networks: Rebuilding Connectivity After Disaster
A new framework leverages high-altitude platforms and drones to rapidly restore communication and provide critical environmental data in the wake of natural disasters.
Predicting Failure: A New Approach to System Lifespan
A novel statistical framework leverages both past performance and real-time data to more accurately forecast when engineering systems will need maintenance or replacement.
The Vanishing Context: When AI Minds Lose Focus
New research reveals a surprising performance drop in large language models as they process longer sequences of text, even when architecturally capable.
Forecasting ED Crowds: Can Machine Learning Predict Hospital Admissions?
New research explores the potential of machine learning models to accurately forecast daily arrivals in emergency departments, offering insights for improved resource allocation.
Seeing Beyond Pixels: Improving Medical Image Segmentation with Frequency Data
![Phi-SegNet employs bi-feature mask formers and attention-guided skip connections to integrate encoder features, then refines segmentation through phase supervision and reverse Fourier attention [latex] \mathcal{R}\mathcal{F} [/latex] modules-a spectral filtering approach designed to sharpen boundary localization despite the inevitable complexities of production deployment.](https://arxiv.org/html/2601.16064v1/Figures/total_architecture.jpg)
A new deep learning framework, Phi-SegNet, boosts the accuracy of medical image analysis by incorporating often-overlooked phase information from the frequency domain.
Predicting Granular Waves with AI: A Faster Path to Material Modeling
Researchers have demonstrated that machine learning can accurately predict the complex propagation of waves through granular materials, offering a significant speedup over traditional simulations.
Catching Starlight’s Last Whisper: A New Supernova Early Warning System
Researchers have developed a novel method to detect the faint neutrino signals emitted before a star collapses, potentially providing hours of advance notice for supernova observations.
Beyond the Context Window: Building Agents with Long-Term Memory
![Aeon’s query latency distribution reveals a system designed for speed-achieving sub-millisecond response for the vast majority of requests [latex] (85\%) [/latex]-but acknowledging the inevitable cost of occasional, longer delays [latex] (up to 2.5ms) [/latex] when cached data is unavailable, a characteristic notably absent in the consistently stable, yet slower, [latex] 1.5ms [/latex] response time of HNSW.](https://arxiv.org/html/2601.15311v1/figures/latency_cdf.png)
New research introduces a system for managing information recall that dramatically extends the coherent reasoning capabilities of artificial intelligence agents.