Research map / Large language models / Language, safety and interpretability
Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs
Qi Li, Runpeng Yu, Haiquan Lu, Xinchao Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use DDM instead of raw confidence in GTA (ablation) | GTA · IRA, CodeAlpaca-20K | AUC: 58.31 -> 65.05 | Improved |
| use DDM instead of raw confidence in GTA (ablation) | GTA · IRA, GSM8K | AUC: 76.79 -> 81.75 | Improved |
| use DDM instead of raw confidence in GTA (ablation) | GTA · CMA, CodeAlpaca-20K | AUC: 96.45 -> 98.94 | Improved |
| use DDM instead of raw confidence in GTA (ablation) | GTA · CCA, CodeAlpaca-20K | AUC: 53.86 -> 62.64 | Improved |
| use GTA instead of Distance attribution on DDM (ablation) | DDM · IRA, GSM8K | AUC: 79.49 -> 81.75 | Improved |
| use GTA instead of Distance attribution on DDM (ablation) | DDM · IRA, CodeAlpaca-20K | AUC: 58.73 -> 65.05 | Improved |
| use GTA instead of Distance attribution on DDM (ablation) | DDM · CMA, GSM8K | AUC: 99.92 -> 99.95 | No clear effect |
| use GTA+DDM instead of Perplexity | GTA+DDM · CMA, GSM8K | AUC: 69.52 -> 99.95 | Improved |
| use GTA+DDM instead of Perplexity | GTA+DDM · CCA, CodeAlpaca-20K | AUC: 40.89 -> 62.64 | Improved |
| zero out gamma and beta effect values in DDM (ablation) | DDM · attribution across 5 token/decoding settings | AUC: performance drops drastically when both beta and gamma are zeroed out | Improved |
| add gamma effect value to DDM (most critical, then beta) (ablation) | DDM · attribution | AUC | Improved |
| replace cell-wise Gaussian with step-wise (Row) Gaussian (ablation) | GTA · attribution | AUC | Hurt |
Closest papers by meaning
- Fast-dLLM++: Fr\'{e}chet Profile Decoding for Faster Diffusion LLM Inference (NeurIPS 2026 · similarity 0.88)
- DAWN: Dependency-Aware Fast Inference for Diffusion LLMs (NeurIPS 2026 · similarity 0.88)
- Confidence-Based Decoding is Provably Efficient for Diffusion Language Models (NeurIPS 2026 · similarity 0.88)
- PACE-dLLM: Elastic Block Decoding via Confidence Cliff Estimation for Diffusion Language Models (NeurIPS 2026 · similarity 0.87)
- d2: Improving Reasoning in Diffusion Language Models via Trajectory Likelihood Estimation (ICML 2026 · similarity 0.87)
- Free Draft-and-Verification: Toward Lossless Parallel Decoding for Diffusion Large Language Models (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
