arXiv subjects

Şeyda Ertekin

Publications and source records attributed to Şeyda Ertekin.

4 recordsLinked to original sources

arXiv2026

ARC-CT: Anatomy-Routed Contrastive Vision-Language Learning for 3D Chest CT

Contrastive vision-language learning uses paired chest CT volumes and radiology reports to learn abnormality classifiers without manually annotated labels. However, two characteristics of chest CT challenge conventional global contrastive learning. First, many critical abnormalities are small or anatomically localized, and pooling an en- tire volume into a single embedding may dilute their visual evidence. Second, the standard contrastive objective treats every other scan in a batch as a negative. Because many chest CTs share abnormalities, this objective incorrectly pushes co-positive pairs apart. We propose Anatomy-Routed Contrastive Learning for 3D Chest CT (ARC-CT), a region-aware framework that addresses these limitations using only la- bels extracted from reports by an LLM, with no manual annotations or bounding boxes. ARC-CT combines three components: (1) an Anato- myQFormer localizing evidence via queries constrained by automatically generated organ masks; (2) a label-Jaccard soft InfoNCE objective in- tegrating the standard one-hot target with the label-set overlap of each pair, which reduces false-negative penalties between studies that share clinical findings; and (3) an organ-level alignment loss connecting mask- pooled visual features to organ-specific report text extracted offline with a large language model. ARC-CT achieves a 0.86 mask-free macro AUC across 18 abnormalities using a compact 3D ResNet-18 backbone. Over- all, ARC-CT outperforms both comparable efficient baselines and sev- eral larger transformer models. Our code and weights are available at https://github.com/arc-ct/arc-ct.

cs.CV↗

arXiv2025

Closing the Gap: Data-Centric Fine-Tuning of Vision Language Models for the Standardized Exam Questions

Multimodal reasoning has become a cornerstone of modern AI research. Standardized exam questions offer a uniquely rigorous testbed for such reasoning, providing structured visual contexts and verifiable answers. While recent progress has largely focused on algorithmic advances such as reinforcement learning (e.g., GRPO, DPO), the data centric foundations of vision language reasoning remain less explored. We show that supervised fine-tuning (SFT) with high-quality data can rival proprietary approaches. To this end, we compile a 161.4 million token multimodal dataset combining textbook question-solution pairs, curriculum aligned diagrams, and contextual materials, and fine-tune Qwen-2.5VL-32B using an optimized reasoning syntax (QMSA). The resulting model achieves 78.6% accuracy, only 1.0% below Gemini 2.0 Flash, on our newly released benchmark YKSUniform, which standardizes 1,854 multimodal exam questions across 309 curriculum topics. Our results reveal that data composition and representational syntax play a decisive role in multimodal reasoning. This work establishes a data centric framework for advancing open weight vision language models, demonstrating that carefully curated and curriculum-grounded multimodal data can elevate supervised fine-tuning to near state-of-the-art performance.

cs.CV↗

arXiv2018

Improving forecasting accuracy of time series data using a new ARIMA-ANN hybrid method and empirical mode decomposition

Many applications in different domains produce large amount of time series data. Making accurate forecasting is critical for many decision makers. Various time series forecasting methods exist which use linear and nonlinear models separately or combination of both. Studies show that combining of linear and nonlinear models can be effective to improve forecasting performance. However, some assumptions that those existing methods make, might restrict their performance in certain situations. We provide a new Autoregressive Integrated Moving Average (ARIMA)-Artificial Neural Network(ANN) hybrid method that work in a more general framework. Experimental results show that strategies for decomposing the original data and for combining linear and nonlinear models throughout the hybridization process are key factors in the forecasting performance of the methods. By using appropriate strategies, our hybrid method can be an effective way to improve forecasting accuracy obtained by traditional hybrid methods and also either of the individual methods used separately.

cs.LG↗

arXiv2015

Reactive point processes: A new approach to predicting power failures in underground electrical systems

Reactive point processes (RPPs) are a new statistical model designed for predicting discrete events in time based on past history. RPPs were developed to handle an important problem within the domain of electrical grid reliability: short-term prediction of electrical grid failures ("manhole events"), including outages, fires, explosions and smoking manholes, which can cause threats to public safety and reliability of electrical service in cities. RPPs incorporate self-exciting, self-regulating and saturating components. The self-excitement occurs as a result of a past event, which causes a temporary rise in vulner ability to future events. The self-regulation occurs as a result of an external inspection which temporarily lowers vulnerability to future events. RPPs can saturate when too many events or inspections occur close together, which ensures that the probability of an event stays within a realistic range. Two of the operational challenges for power companies are (i) making continuous-time failure predictions, and (ii) cost/benefit analysis for decision making and proactive maintenance. RPPs are naturally suited for handling both of these challenges. We use the model to predict power-grid failures in Manhattan over a short-term horizon, and to provide a cost/benefit analysis of different proactive maintenance programs.

stat.AP↗