YoungDSMLKZ at SemEval-2026 Task 13: MIL-UniXcoder with Meta-Stacking and Handcrafted Features for AI-Generated Code Detection

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

Summary

Team YoungDSMLKZ developed a multi-view ensemble framework for 4-class AI-generated code detection, distinguishing between Human, AI, Hybrid, and Adversarial code in long-form repositories. This system achieved 1st place among over 50 teams in SemEval-2026 Task 13 Subtask C, recording a macro F1 score of 0.7855, which was 5.2 points higher than the runner-up. The framework integrates three key components: a Dynamic Multiple Instance Learning (MIL) pipeline using UniXcoder chunks for O(N)-scalable long-context detection, transformer-based meta-stacking combining UniXcoder and ModernBERT, and an XGBoost classifier leveraging over 200 handcrafted stylometric features. Analysis revealed that 62.4% of critical AI-detection signals extend beyond the standard 512-token window, validating the MIL approach's necessity for handling extended contexts.

Key takeaway

For Machine Learning Engineers developing AI-generated code detection systems, you should prioritize long-context analysis and multi-view ensemble approaches. The YoungDSMLKZ framework's success, particularly its MIL pipeline addressing signals beyond 512 tokens, indicates that relying solely on short-context models will yield suboptimal results. Consider integrating transformer-based meta-stacking and handcrafted stylometric features to significantly improve detection accuracy and robustness in realistic, long-form code repositories.

Key insights

The YoungDSMLKZ framework excels at long-context AI-generated code detection by combining MIL, meta-stacking, and stylometric features, outperforming competitors.

Principles

• Long-context signals are crucial for AI code detection.
• Ensemble methods enhance detection accuracy.
• Stylometric features provide strong discriminative power.

Method

The framework employs a Dynamic MIL pipeline over UniXcoder chunks, transformer-based meta-stacking with UniXcoder and ModernBERT, and an XGBoost classifier on 200+ handcrafted stylometric features for 4-class code detection.

In practice

• Implement MIL for code analysis beyond 512 tokens.
• Combine diverse models via meta-stacking.
• Extract stylometric features for code authorship tasks.

Topics

• AI-Generated Code Detection
• Multiple Instance Learning
• UniXcoder
• Transformer Ensembles
• Stylometric Features
• SemEval-2026

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

Related on AIssential

• SpecDetect4ML: Detecting Non-Local ML Code Smells with Code Property Graphs — SpecDetect4AI uses a DSL and static analysis to detect AI-specific code smells with high precision and recall.
• Detecting Functional Memorization in Code Language Models — Code LLMs can functionally memorize logic even when textually dissimilar, requiring advanced auditing.
• SV-Detect: AI-generated Text Detection with Steering Vectors — Steering vectors from frozen LLM hidden states provide a robust, interpretable signal for detecting AI-generated text.
• Guiding Multi-Objective Genetic Programming with Description Length Improves Symbolic Regression Solutions — Description Length (DL) and Fractional Bayes Factor (FBF) effectively prevent overfitting and bloat in symbolic regression when used with multi-objective genetic programming.
• Intro to NLP with spaCy (4): Detecting programming languages — Training a custom spaCy Named Entity Recognition (NER) model automates programming language detection by learning from labeled data.
• CodeMMR: Bridging Natural Language, Code, and Image for Unified Retrieval — Multimodal code retrieval unifies natural language, code, and images to improve code search and generation.

Open in AIssential →

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