cs papers | Gist.Science

Cethraian-Shift is a reproducible medical-imaging research project evaluating the cross-dataset reliability, calibration, label-policy sensitivity, and Grad-CAM++ stability of multi-label chest X-ray classifiers across NIH ChestX-ray14 and VinDr-CXR

The Cethraian-Shift project evaluates the cross-dataset reliability, calibration, label-policy sensitivity, and Grad-CAM++ stability of multi-label chest X-ray classifiers trained on NIH ChestX-ray14 and tested on VinDr-CXR, revealing that discrimination, calibration, and explanation stability behave as distinct technical dimensions while emphasizing that these retrospective findings do not establish clinical utility or deployment readiness.

Mohammed Badhan2026-08-06

💻 computer science

Machine Learning Versus Self-Supervised Transfer Learning for 30-Day Readmission Prediction in Diabetic Patients

This study compares six machine learning models on the UCI Diabetes 130-US Hospitals dataset and finds that tree-based ensembles, particularly XGBoost, outperform both neural networks trained from scratch and those using self-supervised pretraining, achieving a practical discrimination ceiling of approximately 0.675 AUROC for predicting 30-day readmissions in diabetic patients.

Abigail Boatemaa, Baffour Osei, Gabriel Osei Forkuo2026-08-06

💻 computer science

A Hybrid CNN–SPN Framework for Reliable Monitoring and Instability Detection in Hadoop Clusters

This paper proposes a hybrid CNN–SPN framework that integrates a multi-density Stochastic Petri Net with a Convolutional Neural Network to improve the accuracy and interpretability of detecting node instability in Hadoop clusters, achieving superior performance over baseline models on a synthetic dataset.

Walid Ben Mesmia, Zied Trifa, Kamel Barkaoui2026-08-06

💻 computer science

What Does Development Add? Multiple-State Evidence Improves Causal Recovery in Constructed Transformer-Circuit Dossiers

This study demonstrates that providing a language model analyst with multiple sequential states of a constructed transformer circuit, rather than just its final state, significantly improves the accuracy of causal edge recovery and post-intervention predictions, even when all other experimental conditions remain identical.

Zuo Yuchen2026-08-06

💻 computer science

Testing Whether Internal Backdoor Precursors Precede Behavior: A Causal, Power-Matched Protocol

This paper proposes a causal, power-matched protocol to rigorously test whether internal backdoor precursors precede observable malicious behavior by comparing a poisoned model against a matched-null condition using statistically corrected, temporally aligned detectors to avoid false positives from multiple testing and confounding factors.

Zuo Yuchen2026-08-06

💻 computer science

Testing History Dependence Between In-Context and In-Weights Learning

This study demonstrates that while the order of training (approaching a balanced mixture from different directions) creates a transient, measurable difference in the strength of in-context versus in-weights learning mechanisms, it does not result in persistent hysteresis or fundamentally different causal circuit organizations in small transformers.

Zuo Yuchen2026-08-06

💻 computer science

A Critical Period for Compositional Visual Grounding? Controlled Block Order and Causal Attention Interventions in a Small Vision–Language Transformer

This study investigates whether the timing of exposure to relational language affects compositional visual grounding in a small vision–language transformer by manipulating block order and performing causal interventions, ultimately finding no evidence that early exposure provides a performance advantage over late exposure.

Zuo Yuchen2026-08-06

💻 computer science

Does Post-Training Order Change the Mechanism of Safety Alignment? A Controlled Study of Safety DPO and Helpfulness SFT

This controlled study demonstrates that the order of post-training significantly impacts safety alignment mechanisms and outcomes, revealing that training for helpfulness after safety alignment drastically reduces both unsafe refusals and benign over-refusals compared to the reverse order, while also showing that these behavioral shifts are driven by rotating representations in upper layers rather than stable causal mediators.

Zuo Yuchen2026-08-06

💻 computer science

Evidence-Based AI Marking for Secondary Assessment: Alignment, Bias, and Confidence Analysis

This study demonstrates that while an evidence-based Large Language Model shows only moderate alignment with teacher scores and a tendency toward leniency in senior secondary assessments, it serves as a valuable, transparent tool for consistency-oriented moderation and research rather than for fully autonomous high-stakes grading.

Tom Bryden2026-08-06

💻 computer science

Enhanced Gold Mining Optimization with Dual-Layer Information Sharing and Dynamic Control for Multi-Objective Scheduling of Heterogeneous UAV

This paper proposes an Improved Gold Mine Optimization Algorithm with a dual-layer information-sharing framework and dynamic Pareto dominance (IGMO-DP) to effectively solve the multi-objective scheduling problem of heterogeneous UAV swarms by simultaneously minimizing mission time, energy consumption, and load imbalance, demonstrating superior convergence and diversity compared to existing state-of-the-art algorithms.

Siyuan Wei, zhongming lin2026-08-06

← Previous Next →