ASTraNet at SemEval-2026 Task 13: Not All Code Looks the Same: Multi-View Structural and Semantic Detection of Machine-Generated Code

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Expert, medium

Summary

ASTraNet is a system designed to detect machine-generated code, addressing challenges in code quality, security, and authorship verification arising from large language models. Presented at SemEval-2026 Task 13, the system integrates three code-pretrained transformer encoders—CodeT5p-220M, CodeBERT, and UniXcoder—with a structure-first Flow-Augmented AST (FA-AST) encoder, implemented as a Gated Graph Neural Network. This multi-view approach aims to handle unseen programming languages, code generators, and application domains. For Subtask A, ASTraNet's best single model achieved a macro F1 score of 0.559, which improved to 0.643 using a post-competition layered rank-fusion ensemble. In Subtask C, the system officially scored 0.585, further enhanced to 0.652 through a three-stage ensemble that combines neural probabilities with LightGBM-based features and class-priority routing. Its contributions include a language-agnostic structural detector and a meta-learner stacking pipeline for multi-class detection under distribution shift.

Key takeaway

For AI Security Engineers or Machine Learning Engineers tasked with verifying code authorship or quality, ASTraNet's multi-view structural and semantic detection system offers a robust approach. You should consider integrating diverse transformer encoders and graph neural networks for comprehensive code analysis. Employing ensemble strategies, like rank-fusion or meta-learner stacking, can significantly boost detection accuracy, especially when facing unseen languages or code generators, mitigating risks associated with machine-generated code.

Key insights

ASTraNet combines structural and semantic analysis with ensemble methods to detect machine-generated code effectively across diverse contexts.

Principles

• Multi-view analysis improves detection robustness.
• Ensemble methods enhance performance via diversity.
• Language-agnostic structural detection is crucial.

Method

The system combines three transformer encoders (CodeT5p-220M, CodeBERT, UniXcoder) with a Gated Graph Neural Network-based FA-AST encoder. It uses layered rank-fusion for binary classification and a three-stage ensemble with LightGBM for multi-class detection.

In practice

• Combine semantic and structural code analysis.
• Employ ensemble techniques for robust detection.

Topics

• Machine-Generated Code Detection
• Code Quality
• Code Security
• Transformer Encoders
• Graph Neural Networks
• Ensemble Learning

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Security Engineer

Related on AIssential

• Image class translation: visual inspection of class-specific hypotheticals and classification based on translation distance — Image translation networks classify by quantifying alteration needed to fit class-specific hypotheticals, offering interpretability.
• Trailmark turns code into graphs — Code analysis benefits significantly from graph-based reasoning over list-based approaches, especially for security and test quality.
• ASTER: Latent Pseudo-Anomaly Generation for Unsupervised Time-Series Anomaly Detection — ASTER generates latent pseudo-anomalies to train a Transformer-based classifier, enhancing unsupervised time-series anomaly detection.
• Beyond NL2Code: A Structured Survey of Multimodal Code Intelligence — Multimodal Code Intelligence connects visual perception to executable programs for complex programming tasks.
• Sem-Detect: Semantic Level Detection of AI Generated Peer-Reviews — AI-generated peer reviews converge semantically, while human reviews, even LLM-refined, retain unique ideas.
• SHIELD-IDS: Structurally Heterogeneous Ensemble with Integrated Layered Defense for Intrusion Detection Systems — SHIELD-IDS enhances ML-based IDS adversarial robustness through a structurally diverse ensemble and a three-layer black-box defense.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.