Securing the IoT Edge: A New Era of Decentralized Trust
As the Internet of Things expands, this review explores how decentralized security approaches are reshaping trust and resilience at the network edge.
Science
Silent Errors in AI: How a Single Bit Flip Can Compromise Large Language Models
![The analysis reveals that the mean transfer characteristic ratio [latex] TCR [/latex] fluctuates with bit position, demonstrating consistent variation across all five concurrency levels-a pattern underscored by the standard deviation observed across those levels.](https://arxiv.org/html/2604.17249v1/x1.png)
New research reveals a critical vulnerability in how large language models manage memory, potentially allowing subtle data corruption to alter outputs without detection.
Shrinking the AI Footprint: Quantizing Transformers for Efficiency
![The study demonstrates that employing a low-rank approximation-specifically utilizing [latex]MXFP8e4[/latex] quantization for the low-rank branch and activations alongside [latex]MXFP4e2[/latex] for the residual branch-within the LoRaQ framework on the PixArt-ΣΣ architecture demonstrably impacts generative quality, as corroborated by quantitative results presented elsewhere.](https://arxiv.org/html/2604.18117v1/sections/04_experiments/figures/table4_fp8/sample_7/r128.png)
A new method dramatically reduces the computational demands of large AI models without sacrificing performance by intelligently compressing their core components.
Decoding Beyond the Code: Taming Errors in QC-MDPC Systems
A new technique significantly improves error correction in QC-MDPC codes by intelligently addressing errors stemming from near-codewords.
Beyond Ghosts: AI Illuminates Stable Gravity Theories
![The viability of a ghost-free and gauge-invariant theory hinges on a precise relationship between correction coefficients: the Weyl-squared coefficient [latex]\beta_{W^2}[/latex] must remain independent of the kinetic term [latex]X[/latex], consequently demanding that the Gauss-Bonnet coefficient [latex]\beta_{GB}[/latex] exhibit a linear dependence on [latex]X[/latex], thereby illustrating a coupled functional dependency essential for theoretical consistency.](https://arxiv.org/html/2604.16531v1/solutionspacevisualization.png)
New research leverages artificial intelligence to reveal a fundamental link between symmetry and stability in modified gravity theories, paving the way for more reliable cosmological models.
Unmasking Quantum Noise: A Pattern-Based Approach
Researchers have developed a new framework to pinpoint the specific, repeating sources of error plaguing near-term quantum computers.
Beyond Parser Success: Operationalizing Post-Quantum Certificate Assurance
Achieving true security with post-quantum cryptography demands a shift from basic validation to a comprehensive, workflow-driven approach to certificate assurance.
Chasing Pointers, Securing Systems: A New Approach to Verifiable Delay
Researchers have developed a novel sequential work primitive that leverages memory execution and pointer chasing to create robust, latency-bound security functions.
Taming Floating-Point Errors with Category Theory
A new framework and automated tool offer a robust approach to verifying the stability of numerical programs, expanding analysis to previously intractable code.
Unlocking Hyperon Decay: A Lattice QCD Approach
New calculations of key decay parameters using lattice QCD are providing crucial insights into the fundamental forces governing hyperon decay and offering a pathway to search for physics beyond the Standard Model.