Storing 2 bytes of data in your Logitech mouse
Researchers found a way to store 2 bytes of data in the hardware of a Logitech mouse, which could allow devices to securely store small amounts of info without extra memory.
news
Computing
Computer science, software, and information security.
Latest
APEX-Searcher: Augmenting LLMs' Search Capabilities through Agentic Planning and Execution
Researchers developed APEX-Searcher, a system that improves how AI language models search for and use external knowledge to answer complex questions, which could make these models more useful for real-world tasks.
preprint
Wanted: Europe’s Missing Cloud Provider
European telecom companies are pooling efforts to develop a homegrown cloud computing service, reducing reliance on major US providers and giving European businesses a local alternative.
news
TDAD: Test-Driven Agentic Development - Reducing Code Regressions in AI Coding Agents via Graph-Based Impact Analysis
AI coding assistants can introduce bugs that break existing tests, but researchers propose a new "test-driven" approach to track code impact and reduce regressions.
preprint
Beyond AI Psychosis and Sycophancy: Structural Drift as a System-Level Safety Failure
AI safety systems that only check individual messages may miss risks that emerge over time, like concerning behavior patterns, posing system-level threats that could affect many people.
preprint
Erasure Thresholds for Hyperbolic and Semi-Hyperbolic Surface Codes
Researchers developed more efficient quantum error correction codes that can tolerate higher rates of data loss, a key step toward reliable quantum computing.
preprint
The Download: OpenAI’s US military deal, and Grok’s CSAM lawsuit
OpenAI has agreed to give the US military access to its powerful AI technology, raising concerns about potential military applications and the ethical implications of advanced AI in warfare.
news
Floci – A free, open-source local AWS emulator
A new free tool called Floci lets developers run AWS services locally, allowing faster testing and development without cloud costs. This makes it easier for more people to build cloud-based applications.
news
To See is Not to Master: Teaching LLMs to Use Private Libraries for Code Generation
Researchers taught AI language models to generate code that uses private software libraries, enabling them to create more useful applications for developers. This advance could help make AI-generated code more practical and powerful for real-world software projects.
preprint
Structure-Aware Multimodal LLM Framework for Trustworthy Near-Field Beam Prediction
Researchers developed a new AI model that can accurately predict the behavior of light beams in complex environments, which could improve wireless communication technologies like 5G and beyond.
preprint
SpecSteer: Synergizing Local Context and Global Reasoning for Efficient Personalized Generation
Researchers have developed a new AI system that can generate personalized content while preserving user privacy. This could lead to more customized services and apps that respect data privacy.
preprint
Generative AI-assisted Participatory Modeling in Socio-Environmental Planning under Deep Uncertainty
Researchers developed a new AI-powered modeling tool to help communities plan for climate challenges, which could make environmental planning more effective and inclusive for local populations.
preprint
Adaptive Loss-tolerant Syndrome Measurements
Researchers developed a new approach to measure errors in quantum systems that can tolerate qubit losses, which is important for building reliable quantum computers.
preprint
AdaMuS: Adaptive Multi-view Sparsity Learning for Dimensionally Unbalanced Data
Researchers developed a new algorithm to combine data from disparate sources, which could improve machine learning on complex real-world datasets.
preprint
A Dual Certificate Approach to Sparsity in Infinite-Width Shallow Neural Networks
Researchers developed a new machine learning technique that can efficiently train shallow neural networks to be sparse, which could lead to faster and more energy-efficient AI models.
preprint
Professional video editing, right in the browser with WebGPU and WASM
Researchers have developed new web-based video editing tools that let people do professional-quality video editing right in their browsers, without needing to install any software. This could make advanced video editing more accessible to a wider audience.
news
Learning Coordinate-based Convolutional Kernels for Continuous SE(3) Equivariant and Efficient Point Cloud Analysis
Researchers developed a new algorithm that can analyze 3D point cloud data more efficiently by incorporating rigid motion symmetries. This could improve 3D object detection and other applications relying on point cloud data.
preprint
In Focus in this category
Inference in Regression Discontinuity Designs with Clustered Data
Researchers have a new way to analyze data from clustered studies, which can help draw better conclusions about things like policy effects on communities. This could lead to more accurate and impactful findings from real-world studies.
preprint
Improving RCT-Based Treatment Effect Estimation Under Covariate Mismatch via Calibrated Alignment
Researchers developed a new statistical method to improve how treatment effects are estimated from clinical trials, even when the trial participants don't fully match the real-world population. This can lead to more accurate predictions of how treatments will work in practice.
preprint
Starting Off on the Wrong Foot: Pitfalls in Data Preparation
Researchers found that mistakes in data preparation for real-world insurance data can undermine the validity of later analysis, highlighting the importance of careful data cleaning and processing before modeling.
preprint
ELM: A Hybrid Ensemble of Language Models for Automated Tumor Group Classification in Population-Based Cancer Registries
A new AI system can automatically categorize tumor types in cancer registries, saving hundreds of hours of manual labor annually and helping improve cancer care and tracking at the population level.
preprint
ACT-JEPA: Novel Joint-Embedding Predictive Architecture for Efficient Policy Representation Learning
Researchers developed a new technique to learn efficient policy representations without expensive expert demonstrations. This could lead to more accessible AI systems that can better imitate human decision-making.
preprint
PashtoCorp: A 1.25-Billion-Word Corpus, Evaluation Suite, and Reproducible Pipeline for Low-Resource Language Development
Researchers created a massive Pashto language dataset to help improve AI language models for the 60 million people who speak this underrepresented language.
preprint
Self-Conditioned Denoising for Atomistic Representation Learning
Researchers developed a new machine learning technique to better understand materials at the atomic level. This could lead to improved designs for energy storage, electronics and other applications that rely on advanced materials.
preprint
Robustness, Cost, and Attack-Surface Concentration in Phishing Detection
Researchers found that common phishing detection algorithms are vulnerable to manipulation, meaning they may not be as reliable for protecting people from scams as previously thought.
preprint
Quantum linear system algorithm with optimal queries to initial state preparation
Researchers developed a new quantum algorithm that can solve linear systems faster than classical computers by optimizing the initial state preparation. This advance could enable quantum computers to outperform classical ones for certain computational problems.
preprint
Synthetic Control Misconceptions: Recommendations for Practice
Researchers find that a common method for estimating real-world impacts, called synthetic control, often produces misleading results. This is important because policymakers rely on these techniques to evaluate the effectiveness of new programs.
preprint
AdaMuS: Adaptive Multi-view Sparsity Learning for Dimensionally Unbalanced Data
Researchers developed a new algorithm to combine data from disparate sources, which could improve machine learning on complex real-world datasets.
preprint
Generative Control as Optimization: Time Unconditional Flow Matching for Adaptive and Robust Robotic Control
Researchers developed a new robot control method that can adapt to changing situations, improving safety and reliability of robots in complex environments. This could lead to more capable and responsive robots that can better handle real-world challenges.
preprint
NANOZK: Layerwise Zero-Knowledge Proofs for Verifiable Large Language Model Inference
Researchers have developed a way to cryptographically verify the model used in large language model (LLM) queries, preventing providers from substituting cheaper models or using cached responses. This could give users confidence that they are getting the expected AI inference.
preprint
Fast Real-Axis Eliashberg Calculations: Full-bandwidth solutions beyond the constant density of states approximation
Researchers developed a new algorithm to model superconductivity more accurately, which could lead to improved understanding and prediction of real-world superconducting materials.
preprint
Minimum-Action Learning: Energy-Constrained Symbolic Model Selection for Physical Law Identification from Noisy Data
Researchers developed a new AI method to automatically identify physical laws from noisy data, which could help scientists better understand complex natural phenomena.
preprint
PromptHub: Enhancing Multi-Prompt Visual In-Context Learning with Locality-Aware Fusion, Concentration and Alignment
Researchers developed a new AI technique called PromptHub that can more effectively combine visual demonstrations to complete visual tasks, potentially improving real-world AI applications.
preprint