Science – Page 57 – Minority Mindset

Decoding Transformer Chaos: A Spectral Approach to Stable Training

28.02.2026 by qfx

Spectral analysis of a pre-layer normalization transformer reveals that early layers maintain a relatively stable dynamic regime, clustering near the unit circle, while later layers exhibit increasing spectral radius, suggesting a shift toward less constrained and potentially more expressive, but also less stable, representations as information propagates through the network.

A new method analyzes the initial dynamics of transformer layers to predict and prevent the training instabilities that plague these powerful models.

The Rhythm of Learning: How Kernel Spectra Shape Neural Network Training

28.02.2026 by qfx

$The study demonstrates a decomposition of test error into bias and variance components, revealing that the expectation value of the kernel-following a power law of [latex]\Lambda\_{ij}=i^{-3/2}\delta\_{ij}[/latex]-dictates the trade-off between these error sources, as observed through simulations employing a time step of [latex]\mathrm{d}t=10^{-4}[/latex] and averaged over [latex]10^{5}[/latex] realizations with parameters [latex]\beta=10[/latex] and [latex]g\beta=10^{3}[/latex] at an interpolation threshold of P=N=102, contrasted with theoretical calculations utilizing [latex]\mathrm{d}t=10^{-2}[/latex].$

New research reveals the interplay between kernel structure and training dynamics, offering insights into why and how neural networks generalize effectively.

Silent Sabotage: Hacking Vehicle Recognition with Subtle Sound

28.02.2026 by qfx

New research reveals that acoustic vehicle classification systems are surprisingly vulnerable to data poisoning attacks, even with minimal data corruption.

Unlocking Dynamics: A New Approach to State-Space Models

28.02.2026 by qfx

Researchers have developed a constrained optimization framework and a novel model, the Extended Kalman VAE, to significantly improve the learning of complex, dynamic systems.

Predicting Cloud Demand with AI: A New Topology-Aware Approach

28.02.2026 by qfx

A novel artificial intelligence framework leverages service dependencies and multi-granularity data to significantly improve load forecasting in dynamic cloud native platforms.

Predicting the Rise of Superbugs: A Data-Driven Approach

28.02.2026 by qfx

XGBoost’s predictive accuracy concentrates within a narrow margin of error for most observations, yet significant deviations consistently arise when analyzing high-resistance data, suggesting a limitation in the model’s capacity to extrapolate beyond commonly observed conditions.

New research harnesses global surveillance data and machine learning to forecast antimicrobial resistance trends and inform public health strategies.

Mapping the Flow: A Smarter Approach to Traffic Forecasting

28.02.2026 by qfx

Performance comparisons across varying traffic scales reveal that reproduced models-along with those cited from previous work-demonstrate limitations on hardware with 48 GB of memory, as indicated by out-of-memory errors denoted by a hyphen, while the bolded values highlight the best-performing configurations within those constraints.

A new spatio-temporal network leverages positional awareness and temporal attention to dramatically improve the accuracy and efficiency of large-scale traffic prediction.

The Search for Truth Online: A China-Focused Study

28.02.2026 by qfx

Across multiple search engines-Baidu, Bing, and Sogou-large language models including DeepSeek, Qwen, and LLaMA demonstrated varying predictive accuracy, with confidence intervals established through bootstrapping, and performance generally clustered around an overall mean as indicated by the red dashed line.

A new analysis reveals the varying accuracy of search engines, large language models, and AI-powered overviews in delivering factual information to Chinese web users.

Code Under Scrutiny: AI Spots Vulnerabilities with Growing Accuracy

28.02.2026 by qfx

A new approach leveraging deep learning is significantly improving the automated detection of security flaws in source code.

Decoding Attacks to Uncover Hidden Vulnerabilities

28.02.2026 by qfx

New research shows how analyzing the language of cyberattacks can proactively identify software flaws before they are exploited.