cs papers | Gist.Science

When Does Context Help Machine Vision? Decomposing the Risk and Reliability of Contextual Conditioning in Vision-Language Models

This paper systematically analyzes contextual conditioning in vision-language models across diverse architectures and datasets, revealing that while domain-level prompting is consistently beneficial, full per-image contextual conditioning carries high variance and overfitting risks, with its utility primarily driven by model scale and baseline accuracy.

Alaa Alahmadi2026-07-13

💻 computer science

Mining AI-Assisted Course Design Workflows at Production Scale

This paper presents the first production-scale study of AI-assisted course design by mining a privacy-preserving dataset from CourseFactory to extract four workflow surfaces, demonstrating that structural-quality triage and item-to-assignment routing models significantly improve review efficiency and accuracy while confirming that withheld prompt text adds no measurable signal.

Aleksandr Volkov, Taras Pustovoy, Dmitriy Istomin, Viacheslav Istomin, Tatiana Otbetkina2026-07-13

💻 computer science

Recovery Dynamics and Persistent Structural Excursion in Disrupted World Models

This study demonstrates that while successful and failed trajectories in a disrupted world model exhibit distinct recovery dynamics—such as lower path entropy and fewer structural excursions—these full-trajectory patterns cannot be reliably predicted from early prefixes, thereby distinguishing the challenges of regime detection, recovery monitoring, and individual rollout prediction.

Jeffery Scott Allbright2026-07-13

💻 computer science

H. Dilpriya's Momentum (HDM) : A Multi-Strategy Gradient-Aligned Optimizer with Adaptive Per-Parameter Corrections, Cosine-Annealed Scheduling, and Convergence Guarantees for Deep Neural Networks

This paper introduces H. Dilpriya's Momentum (HDM), a novel multi-strategy optimizer that combines adaptive per-parameter corrections with cosine-annealed scheduling to achieve rigorous convergence guarantees and state-of-the-art gradient alignment, demonstrating superior performance on both ill-conditioned synthetic problems and deep learning benchmarks like MNIST and CIFAR-10.

Janaka Ishan Senarathna2026-07-13

💻 computer science

Multi-Tenant Edge-Cloud Hybrid LLM Serving using Speculative Decoding and Quantization

This paper proposes a multi-tenant edge-cloud hybrid LLM serving architecture that combines W4A16-quantized speculative decoding with asynchronous communication and a multi-tenant verification orchestrator to simultaneously address privacy, latency, and resource utilization challenges.

Jui-Yu Lin2026-07-13

💻 computer science

Learning Safe Behaviour via Justified Human Preferences and Hypothetical Queries

This paper introduces JPAL-HA, a safe learning model for safety-critical environments that leverages human action justifications to generate hypothetical queries, thereby significantly improving safety and sample efficiency while reducing human training burden.

Ilias Kazantzidis, Timothy J. Norman, Yali Du, Christopher T. Freeman2026-07-13

💻 computer science

Understanding Behavioural Risk Signals in Socio-Technical Work Systems Using Explainable Machine Learning

This study demonstrates that using explainable machine learning to analyze item-level safety climate data reveals nonlinear, threshold-dependent behavioural risk patterns driven by management responsiveness, training, and experience, challenging traditional linear aggregation models and offering a more precise basis for proactive accident prevention.

Omid Akbarzadeh, Parisa Moshashaei, Rasoul Ahmadpour, Mohammed Qadir Ali, Seyed Shamseddin Alizadeh2026-07-13

💻 computer science

ℓ0-Regularized Quadratic Surface Support Vector Machines

This paper proposes a sparse ℓ0-regularized quadratic surface support vector machine (QSVM) to address overfitting and interpretability issues in kernel-free nonlinear classification, introducing a penalty decomposition algorithm with provable optimality and convergence guarantees that demonstrates competitive performance and sparsity on both benchmark and real-world credit datasets.

Ahmad Mousavi, Ramin Zandvakili, Zheming Gao2026-07-13

💻 computer science

AI-Driven Multimodal Communication Framework for Inclusive Classrooms: A PRISMA-Based Systematic Review for Deaf Learners

This paper presents a PRISMA-based systematic review revealing the limitations of current auditory-centric solutions for deaf learners and proposes an AI-driven multimodal communication framework integrating speech recognition, natural language processing, and sign language generation to enable real-time visual support and achieve meaningful inclusion in classrooms.

shahir vk2026-07-13

💻 computer science

Empty scaffold allocation bias in molecular distribution shift evaluation

This paper reveals that standard scaffold-based evaluation methods introduce a hidden allocation bias by incorrectly treating chemically heterogeneous "no-scaffold" molecules as a single unit, which distorts generalization metrics across common benchmarks, and proposes an "Empty-Scaffold-Aware" repair to restore balanced representation and accurate performance assessment.

Tao Song, Yong Wang, Shuang Wang, Xun Wang2026-07-13

← Previous Next →