Hijacking the Smart Home: How Attackers Blind IoT Security
New research reveals how carefully crafted network traffic can evade machine learning-based intrusion detection systems protecting Internet of Things devices.
Science
Seeing Beyond the Norm: AI Spots Flaws on the Factory Floor
![The study demonstrates a comparison between direct feature reconstruction and template-based feature aggregation, wherein reconstructed features [latex]Rec.[/latex] are contrasted with corresponding anomaly maps [latex]Ano.[/latex] to evaluate performance.](https://arxiv.org/html/2603.22874v1/pic/global-defect.png)
A new approach to industrial anomaly detection uses the power of vision transformers to reconstruct expected patterns and highlight deviations, improving quality control and reducing defects.
News Flash: Do Circuit Breakers Distort Market Signals?
New research examines whether automated trading halts impact how quickly and accurately financial markets incorporate important news.
The Ghost in the Machine: Why AI Writing Detectors Keep Failing
![The analysis of feature impact, conducted on both human-authored and artificially generated samples, reveals discernible patterns in their respective contributions, suggesting that [latex] SHAP [/latex] values can effectively differentiate between these two data sources based on the relative importance of underlying features.](https://arxiv.org/html/2603.23146v1/Figures/ai-generated_sample_bar_plot.png)
A new analysis reveals that current methods for identifying AI-generated text often mistake stylistic quirks for genuine signs of machine authorship.
AI Portfolio Pilots: Building Smarter Investment Strategies
A new framework utilizes artificial intelligence to analyze vast datasets and construct portfolios that consistently outperform traditional methods.
The AI Model Marketplace: Finding Profit in Price Gaps
A new analysis reveals opportunities to profit from price discrepancies between artificial intelligence models, driving a dynamic and potentially efficient market.
Seeing Beyond the Noise: A New Approach to Detecting the Unknown
![SPECTRE-G2 operates on the principle that anomaly detection benefits from diverse perspectives, employing parallel processing through spectral-normalized Gaussian encoders and a PlainNet to extract eight complementary signals, which are then refined via validation percentiles and out-of-distribution correction before an adaptive fusion-prioritized by validation AUROC-averages the most discriminative signals into a final anomaly score [latex] S(\mathbf{x}) [/latex], with an optional causal signal incorporated for tabular data to further enhance sensitivity.](https://arxiv.org/html/2603.21160v1/arch.png)
Researchers have developed a novel anomaly detection system that leverages multiple data signals and uncertainty measurements to identify truly unexpected events.
Decoding Market Signals: Beyond Sentiment’s Surface
![Despite the expected temporal decay of market sentiment, analysis reveals that NextEra’s lag-2 coefficient-measuring sentiment’s influence two periods prior-uniquely withstands rigorous refutation across multiple tests, suggesting a sustained, albeit delayed, predictive power not observed at other lags [latex] (0, 1, 3) [/latex].](https://arxiv.org/html/2603.21473v1/fig_lagprofile.png)
New research reveals a method for separating genuine connections between public opinion and energy market returns from misleading correlations.
Predicting the Unexpected: Forecasting Long-Term Anomalies in Time Series
![The system accurately forecasts outlier events in the Beijing bivariate dataset, demonstrated by its prediction of a temperature spike at time [latex]T=10597[/latex] and a subsequent rise in the pressure score at [latex]T=10607[/latex], confirming its ability to anticipate anomalous behavior.](https://arxiv.org/html/2603.20993v1/x6.png)
A new framework learns the patterns of detected anomalies to forecast outliers far beyond immediate deviations in time series data.
Uncovering Price Anomalies with AI Reasoning
A new framework uses artificial intelligence to break down complex price checks into understandable steps, mirroring human audit processes.