2026· SemEval@ACL· pp. 2233-2237· 0 citations· 8 references
Computer Science
TL;DR
The submission to SemEval-2026 Task 13 (Orel et al., 2026), addressing binary detection, generator attribution, and hybrid/adversarial author-ship classification of machine-generated code of machine-generated code (MGC).
Abstract
We describe our submission to SemEval-2026 Task 13 (Orel et al., 2026), addressing binary detection (Subtask A), generator attribution (Subtask B), and hybrid/adversarial author-ship classification (Subtask C) of machine-generated code (MGC). For Subtask A, we fine-tune two CodeBERT (Feng et al., 2020) models with complementary sampling strategies and apply percentile-based post-hoc calibration, improving Macro-F1 from 0.47 to 0.56 without additional training. For Subtask B, we combine TF-IDF n-grams, frozen Code-BERT embeddings, and language features with XGBoost (Chen and Guestrin, 2016), us-ing synthetic augmentation and class weighting to handle an 11-class dataset skewed 88% toward the human class, achieving Macro-F1 of 0.289. For Subtask C, we fine-tune a Code-BERT classifier for four-way authorship classification, achieving Macro-F1 of 0.49. Our results highlight the importance of probability calibration for binary detection and class balancing for multi-class attribution.
This study investigates response-only adversarial classification of LLM outputs through a comparative evaluation of lexical, contrastive, and contextual representations and provides a realistic benchmark for response-only adversarial detection and clarify the limits of contrastive representations when prompt information is absent.
A. Nasution, Ahmet Emre Ergün, Aytu˘g Onan et al.· IEEE Access· 0 citations
Results show that security assessment of self-supervised distillation should include executable training logic in addition to model weights and visible data.
Da Xiao, Tongke Fan, Ning Dong et al.· Electronics· 0 citations
A self-initiated investigation of purification-based adversarial detection, comparing three families of detection signals across six detectors that share a CLIP ViT-L/14 backbone finds that raw $|\Delta \text{logit}|$ under median-3 purification, applied through the EFFORT detector, separates adversarial inputs from clean inputs with AUROC 0.81-0.98.
Junghyun Kim, Seunghyun Kim, Ji-myung Woo· 0 citations
The original ALPHA benchmark introduced a taxonomy-aware penalty for evaluating CWE-level vulnerability prediction in Python and proposed that the penalty could theoretically also serve as a training signal. This paper tests that proposal empirically, comparing three delivery mechanisms–supervised fine-tuning, a dual-head classification loss, and reinforcement learning with a dense reward derived from the normalised penalty. We find that supervised approaches consistently regress below the zero-shot baseline under distribution shift, while GRPO succeeds and generalises across independent training seeds. Using a default, non-test-informed configuration, GRPO reduces the mean cumulative ALPHA penalty of Qwen2.5-Coder-7B-Instruct on Security Hardening and Adversarial Testing (SVEN) by 13.5%. A further test-selected configuration reduces the mean penalty by 22.9% (27.9% in the best-observed run), reaching statistical parity with its 4.5× larger zero-shot teacher. Our experiments show that reinforcement learning outperforms the alternative delivery mechanisms we test. We conclude that delivery mechanism, rather than the specific hierarchical structure of the reward, is the primary variable—though whether taxonomy-aware structure contributes further remains an open question.
Muntasir Adnan, Manile Srun, Carlos C. N. Kuhn· Machine Learning and Knowled...· 0 citations
CoGate is proposed, a confidence-gated co-decoding approach that controls the expert's influence on the co-decoding process based on its confidence, and outperforms existing co-decoding methods (CoSec+) across multiple benchmarks.
Minghao Hu, Lannan Luo, Allen G. Roush et al.· 0 citations