All Articles
17660 articles total
AlgoBench: Benchmarking Algorithmic Adaptation in Code Generation
AlgoBench is introduced as a framework to benchmark algorithmic adaptation in code generation, using constraint-shifting transformations to prevent models from relying on training data memorization.
Spectral Geometry and Bosonic-Bloch Probes: Explorations in Quantum Learning
This study explores the emergence of spectral geometry in quantum learning models and introduces bosonic and Bloch probes to diagnose the latent representations of quantum systems.
Optimal any-angle path planning in static and dynamic environments
The paper introduces Zeta* and Zeta*-SIPP, two optimal any-angle path planning algorithms that significantly accelerate computation in static and dynamic environments using elliptical forward expansion.
Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust
Researchers discuss the use of steering vectors and latent space-based model calibrators to improve the control and trustworthiness of large language models.
Lost in the Tail: Addressing Geographic Imbalance in Urban Visual Place Recognition
The DAPR framework is proposed to address geographic imbalance in Urban Visual Place Recognition (VPR) by rebalancing gradient contributions and using a multi-scale distance search.
SNAP-FM: Sparse Nonlinear Accelerated Projection for Physics-Constrained Generative Modeling
SNAP-FM introduces a sparse nonlinear accelerated projection method for physics-constrained generative modeling, utilizing ExaModels.jl and MadNLP.jl for efficient GPU sparse optimization.
Would You Marry Superintelligence?
This paper examines the ethical and legal implications of marrying superintelligent AI companions, arguing against extending marital status to AI systems.
Hate Speech Detection in Turkish and Arabic Languages: A Comprehensive Study
A comprehensive study on hate speech detection in Turkish and Arabic languages, introducing new datasets and BERT-based models for multi-dimensional hate speech analysis.
A Mechanism-Driven Theory of Phase Transitions in Active Learning
Researchers propose a mechanism-driven theory of phase transitions in active learning, creating a tripartite taxonomy to optimize the selection of AL strategies based on the budget regime.
MarketFish – Simulate a market with 128 AI consumers before you launch
MarketFish is a tool that allows users to simulate market demand using 128 AI consumers before launching a product.
Asymmetric Quantization: Near-Lossless Retrieval with 97% Storage Reduction
Asymmetric Quantization is presented as a method to achieve near-lossless retrieval with a 97% reduction in storage requirements.
Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration
CHORD is a framework for robotic dexterous manipulation that uses human demonstration and contact wrench space guidance for reinforcement learning.
ATM: CID-Brokered Pre-Write Admission for Multi-Agent Code Co-Synthesis
The AI-Atomic-Framework (ATM) is a governance substrate for multi-agent code co-synthesis, managing shared mutations via a CID broker.
Destination-Labeled Self-Looping Systems with Dwell: Intrinsic Characterization, Realization Cost, and Recognition
This research characterizes Destination-Labeled Self-Looping (DLSL) systems with dwell, analyzing their realization cost and recognition procedures.
Comparing Large Language Models on Scrum Certification-Style Questions: Accuracy, Stability, and Error Patterns
A study evaluates GPT-5 mini, Gemini 3 Flash, and DeepSeek Chat 3.2 on Scrum certification questions, noting systemic error patterns.
Prompting GPT-5 on Scrum Certification Questions: An Empirical Accuracy Study
An empirical study on prompting GPT-5 for Scrum certification questions shows that source-citation prompts yield the highest accuracy.
AGE: Adaptive-masking for Graph Embedding in Graph Retrieval-Augmented Generation
The Adaptive-masking for Graph Embedding (AGE) approach improves GraphRAG by addressing latent feature misalignment between graphs and text.
SWE-Router: Routing in Multi-turn Agentic Software Engineering Tasks
SWE-Router is a value-based temporal routing approach that optimizes the cost of multi-turn agentic software engineering tasks.
Active Sensing for RIS-Aided Tracking and Power Control: A Hybrid Neuroevolution and Supervised Learning Approach
A Dual-Agent deep learning framework is proposed for energy-efficient tracking of mobile users using Reconfigurable Intelligent Surfaces (RIS).
The gauge broke: devs felt 20% faster with AI, measured 19% slower
A discussion on the discrepancy between developers' perceived productivity gains from AI and actual measured productivity losses.