Publications

For citation metrics and automatically indexed versions, see Google Scholar. Workshop papers are labelled separately from main-conference publications. An asterisk is not used for contribution because author-contribution statements are available for only some manuscripts.

Peer-reviewed conference publications

Beyond Major Floods: Deep Learning for Detecting Shallow Water Inundation in Agricultural Areas

Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz.

29th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES 2025). Procedia Computer Science 270, 301-310.

Ten deep-learning segmentation models for three-class Sentinel-1 SAR mapping; ResNet-UNet matched DeepLabv3+ with lower compute and memory requirements.

Workshop publications

Self-Reports Do Not Identify Self-Models: An Identifiability Test for Counterfactual Reports

Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz.

Oral presentation, Philosophy of Machine Learning Workshop at ICML 2026.

Environment-shift tests across three open models show that correct counterfactual reports do not by themselves identify a self-model bound to the intervention.

Decoded but Unused: Instruction Tuning Routes Moral Framing into the Judgment Readout

Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz.

Poster, ICML 2026 Workshop on Mechanistic Interpretability.

A matched pretrained and instruction-tuned audit separates a representation that is decodable from one that is causally routed into judgment.

A Path Already Walked: On Inheriting Network-Neuroscience Tools for Mechanistic Interpretability

Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz.

Virtual poster, ICML 2026 Workshop on Mechanistic Interpretability.

A position paper specifying the graph objects, projections, null models, and failure criteria needed for a disciplined transfer of network-neuroscience tools.

Architecture Without Architects: How AI Coding Agents Shape Software Architecture

Phongsakon Mark Konrad, Tim Lukas Adam, Raffaele Terrenzi, Serkan Ayvaz.

SAGAI Workshop at IEEE ICSA 2026.

Five mechanisms and six prompt-architecture coupling patterns describe how coding agents make implicit architectural choices.

CAKE: Cloud Architecture Knowledge Evaluation of Large Language Models

Tim Lukas Adam, Phongsakon Mark Konrad, Raffaele Terrenzi, Finn Gunnar Lukas, Rukiye Yilmaz, Kamil Sierszecki, Serkan Ayvaz.

KDA-AI Workshop at IEEE ICSA 2026.

A 188-question, expert-validated benchmark evaluates 22 model configurations across knowledge levels and response formats.

A Reference Architecture for Agentic Hybrid Retrieval in Dataset Search

Raffaele Terrenzi, Phongsakon Mark Konrad, Tim Lukas Adam, Serkan Ayvaz.

SAML Workshop at IEEE ICSA 2026.

An auditable Plan-Retrieve-Evaluate architecture combines BM25, dense retrieval, reciprocal-rank fusion, and offline metadata augmentation.

Preprints and manuscripts

Coverage Debt: Pricing the Conformal Coverage Gap Through a Real Electricity-Market Redesign

Phongsakon Mark Konrad, Tim Lukas Adam, Serkan Ayvaz.

Preprint, 2026. Extends the BSc thesis.

Across eight European bidding zones, a static conformal calibration loses coverage with volatility and creates a matching decision-level loss-floor breach; refreshed calibration restores the guarantee.

Documented contribution: conceptualization, methodology, software, formal analysis, data curation, visualization, and original draft.

Challenges in Deep Learning-Based Small Organ Segmentation: A Benchmarking Perspective for Medical Research with Limited Datasets

Phongsakon Mark Konrad, Andrei-Alexandru Popa, Yaser Sabzehmeidani, Liang Zhong, Madhulika Tripathy, Andrei Constantinescu, Elisa A. Liehn, Serkan Ayvaz.

Preprint, 2025.

Ten models, ablations, bootstrap uncertainty, and independent distribution-shift data show that architecture rankings are unstable in very small cardiovascular histology datasets.

Documented contribution: conceptualization, data curation, formal analysis, methodology, software, visualization, validation, and original draft.

How Much of Non-Destructive Fruit-Ripeness Assessment Is Real? A Confound-Controlled, Low-Cost Visible Hyperspectral Study Across Five Fruit

Phongsakon Mark Konrad, Casper Kunstmann-Olsen, Jacek Fiutowski, Serkan Ayvaz.

Preprint, 2026.

A species-only baseline separates genuine spectral skill from fruit recognition: visible spectra retain a strong within-species ripeness signal, while pooled firmness is heavily confounded.

Documented contribution: conceptualization, methodology, software, formal analysis, investigation, data curation, visualization, and original draft.

Acceptance Cards: A Four-Diagnostic Standard for Safe Fine-Tuning Defense Claims

Phongsakon Mark Konrad, Toygar Tanyel, Serkan Ayvaz.

Preprint, 2026.

A claim-specific audit checks statistical reliability, fresh-subject generalization, mechanism alignment, and transfer before upgrading a held-out improvement to a full defense claim.

The Open-Box Fallacy: Why AI Deployment Needs a Calibrated Verification Regime

Phongsakon Mark Konrad, Tim Lukas Adam, Anders C. H. Merrild, Riccardo de Rosa, Raffaele Terrenzi, Toygar Tanyel, Serkan Ayvaz.

Position paper, preprint, 2026.

Argues that mechanistic evidence should be one input to use-specific deployment governance and proposes a six-component Verification Coverage standard.