Tracking System Health with Geometric Drift
![The study demonstrates a nuanced relationship between feature extraction (F), reliability (R), and operational toil (O) within a simplex, revealing that a Euclidean alarm can trigger even when the F/R ratio remains constant, while conversely, no alarm may sound despite an unsafe F/R ratio exceeding 1.5, indicating the limitations of a simple Euclidean threshold for accurately assessing system state when [latex] F = R [/latex] or [latex] F = 1.5R [/latex].](https://arxiv.org/html/2602.05483v1/x1.png)
A new approach leverages compositional data analysis to monitor evolving software systems and maintain reliable observability.
Science
Beyond Trends: A New Approach to Time Series Forecasting
![Electricity load forecasting performance-measured by metrics including mean squared error (MSE), spectral divergence, autocorrelation MSE, and maximum mean discrepancy-is evaluated across varying context lengths [latex] \ell [/latex], demonstrating the comparative efficacy of LISA, ALSA, and PTST models in predicting energy demand with established statistical rigor.](https://arxiv.org/html/2602.04906v1/figures/electricity_mse.png)
Researchers have developed a novel framework that combines geometric modeling with in-context learning to achieve more accurate and adaptable predictions for complex time series data.
Beyond Single Signals: A New Era for Time Series Analysis
Researchers are harnessing the power of multimodal data and large-scale pretraining to dramatically improve the accuracy and robustness of time series forecasting and anomaly detection.
When Bigger Data Isn’t Better: Scaling Challenges in Materials Science
New research reveals that performance gains from increasing datasets aren’t guaranteed in materials modeling, challenging established scaling laws.
Tracing the Source: Understanding Information Flow in AI Systems
A new framework dissects how AI systems like those powered by retrieval-augmented generation arrive at their answers, offering unprecedented insight into their reasoning.
Beyond Summation: Neural Networks Learn to Aggregate Sets More Effectively
A new approach to set function learning, called Quasi-Arithmetic Neural Networks, boosts expressiveness and transferability by moving beyond simple summation of set elements.
Are AI Explanations Trustworthy?
New research reveals that many popular methods for explaining AI decisions produce unstable and unreliable results, raising concerns about their practical use.
When Models Break: Testing the Limits of Hydrological Prediction
New research reveals how vulnerable common hydrological models are to subtle data manipulations, and surprisingly, which model proves more resilient.
Forecasting the Future: A New Approach to Time Series Analysis
![DecompSSM offers a forecasting approach built upon decomposition into trend, seasonal, and residual components, enhanced by auxiliary objectives promoting orthogonality and reconstruction, and leverages a Gated-Time State Space Model (GT-SSM) incorporating an input-dependent Adaptive Step Predictor-an architecture derived from S5 [smith\_s5\_2023]-to navigate the inherent decay of predictive systems.](https://arxiv.org/html/2602.05389v1/figs/model.png)
Researchers have developed a novel state space model that breaks down complex time series data into core components, leading to improved forecasting accuracy.
Predicting Coastal ‘Dead Zones’ with AI
New research shows artificial intelligence models can accurately forecast daily hypoxia levels in coastal waters, offering critical insights for marine ecosystem management.