Rhythmic Learning: How Brain-Inspired Oscillations Unlock AI Planning
![Memory capacity benchmarks reveal that the Memory-Hierarchical Network (MHN) maintains perfect recall up to a pattern density of [latex]P=N[/latex], significantly exceeding the performance of the Phasor-graph memory-which falls below 95% recall near the classical Hopfield bound of [latex]0.138N[/latex]-and demonstrating a substantial advantage over the Echo State Network, which rapidly degrades in associative recall performance due to its architectural limitations.](https://arxiv.org/html/2601.04362v1/x8.png)
A new approach to reinforcement learning uses the principles of brain rhythms and sleep-like stages to enable agents to learn complex tasks and improve generalization.
Science
Outsmarting the Hack: Fortifying Language Models Against Malicious Commands
New research details a powerful defense against prompt injection attacks, leveraging synthetic data and enhanced reasoning capabilities to protect large language models.
Untangling Knots with State Integrals
New research explores a powerful mathematical technique for calculating knot invariants and gaining insights into their asymptotic behavior.
Spinning Stars: A New Code for Realistic Neutron Star Simulations
Researchers have developed a new computational suite to model the complex interiors and dynamics of rapidly rotating neutron stars with unprecedented accuracy.
Streaming 3D Worlds: A New Codec for Gaussian Splatting
![The research demonstrates a performance trade-off, assessed via an R-D curve on the Bicycle scene from the MipNeRF360 dataset, wherein varying [latex]\lambda_{ssim}[/latex] values-ranging from 0.1 to 0.4-influences the method’s results when benchmarked against PCGS and GoDE.](https://arxiv.org/html/2601.04348v1/images/R-D_Curve.png)
Researchers have developed a progressive codec that leverages spatial context and residual quantization to enable efficient streaming of high-quality 3D Gaussian Splatting scenes.
Stealth Attacks on AI’s Digital Watermarks
New research demonstrates how to subtly dismantle techniques used to identify the origins of large language model outputs, potentially undermining intellectual property protections.
Fortifying Language Models Against Rogue Prompts
A new approach to defending large language models from adversarial attacks prioritizes efficiency and reliability for real-world deployment.
Cosmic Strings Snap: Early Universe Defects Face Their Breaking Point
![The study demonstrates that gravitational wave spectra from metastable cosmic string networks-shaped by parameters including [latex]\kappa = 485[/latex], [latex]G_N\mu = 10^{-7}[/latex], and [latex]\Gamma_g = 50[/latex]-exhibit distinct characteristics dependent on network formation and early-time breaking mechanisms, with finite-temperature restoration and inflationary scenarios yielding spectra that either enhance decay due to thermal effects or require significantly larger κ values-approximately 2000-to align with observed pulsar timing array signals, as indicated by the location of [latex]f_{\rm low}[/latex] defined in Eq. (48).](https://arxiv.org/html/2601.04320v1/x11.png)
New research reveals how metastable cosmic strings-topological defects formed in the early universe-fracture due to finite temperature effects and quantum fluctuations.
Taming Loopy Belief Propagation with Geometric Principles
A new framework stabilizes inference in complex graphical models by leveraging concepts from descent theory and holonomy to address inconsistencies arising from cycles.
Unlocking Schur Positivity: A New Proof for a Long-Standing Conjecture
Researchers have confirmed a key hypothesis concerning the non-negativity of certain polynomial coefficients, resolving a problem that has challenged mathematicians for years.