发表机构
Guangdong Polytechnic Normal University（广东技术师范大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出MAD-Guard，在Qwen3-VL-8B上对比自回归生成与直接决策接口，发现直接头可大幅降低延迟和校准误差，并提升多任务归因性能。
AI 中文摘要
多模态基础模型何时应生成令牌，何时应直接输出决策？我们提出了MAD-Guard，一项针对封闭式多模态取证任务的输出决策接口的受控研究。一旦计算出多模态表示，对于输入复杂度高但输出熵低的封闭式取证决策，自回归生成是否必要？在匹配的Qwen3-VL-8B骨干网络、2,400个FakeClue训练样本以及华为昇腾910C NPU上的LoRA预算（r=16, α=32）下，我们评估了一系列决策接口（AR-SFT [生成] → Logit Slice → Binary Direct Head → +choice → +act → CLM-Head），并将延迟分解为骨干表示（53.12毫秒）、151,643路词汇投影（+85.04毫秒 → 138.16毫秒）和解码（+248.26毫秒 → 386.42毫秒）。在1对1二元监督（L_BCE）下，Binary Direct Head将延迟降低2.60倍至7.27倍（53.12毫秒），并将校准误差降低1.88倍（ECE = 0.0450 vs. 0.0845），但以-1.80%的准确率权衡（93.10% vs. 94.90%；0.9795 vs. 0.9871 ROC-AUC）为代价，源于放弃令牌先验。相对于AR-SFT的增益要么来自多任务归因和不确定性门控（+choice+act：96.44%准确率，0.9940 ROC-AUC，0.0187 ECE，53.71毫秒），要么来自分解的对比头（CLM-Head：96.55%二元和96.44%多任务准确率，0.0166 ECE，98.79% 7类归因，54.42毫秒），在不进行令牌解码的情况下保留语义先验。在来自五个基准的5,000张样本外图像上，我们的框架在合成、伪装和文档伪造方面表现出色（GenImage 96.44%，Chameleon 97.73%，Doc 91.84%），而在压缩人脸操纵上显示出明显边界（FF++ ROC-AUC = 0.5913）。
英文摘要
When should multimodal foundation models generate tokens, and when should they directly output a decision? We present MAD-Guard, a controlled study of output-decision interfaces for closed multimodal forensic tasks. Once a multimodal representation is computed, is autoregressive generation necessary for closed forensic decisions with high input complexity but low output entropy? Under a matched Qwen3-VL-8B backbone, 2,400 FakeClue training samples, and LoRA budget ($r=16, α=32$) on Huawei Ascend 910C NPUs, we evaluate a progression of decision interfaces (AR-SFT [generate] $\to$ Logit Slice $\to$ Binary Direct Head $\to$ +choice $\to$ +act $\to$ CLM-Head) and decompose latency into backbone representation (53.12 ms), 151,643-way vocabulary projection (+85.04 ms $\to$ 138.16 ms), and decoding (+248.26 ms $\to$ 386.42 ms). Under 1-to-1 binary supervision ($\mathcal{L}_{\mathrm{BCE}}$), a Binary Direct Head cuts latency by $2.60\times$-$7.27\times$ (53.12 ms) and lowers calibration error by $1.88\times$ (ECE = 0.0450 vs. 0.0845), with a -1.80% accuracy trade-off (93.10% vs. 94.90%; 0.9795 vs. 0.9871 ROC-AUC) from forfeiting token priors. Gains above AR-SFT arise either from multi-task attribution and uncertainty gating (+choice+act: 96.44% accuracy, 0.9940 ROC-AUC, 0.0187 ECE at 53.71 ms) or from a disaggregated contrastive head (CLM-Head: 96.55% binary and 96.44% multi-task accuracy, 0.0166 ECE, 98.79% 7-class attribution at 54.42 ms) retaining semantic priors without token decoding. Across 5,000 out-of-sample images from five benchmarks, our framework excels on synthetic, camouflage, and document forgeries (96.44% GenImage, 97.73% Chameleon, 91.84% Doc) while showing a clear boundary on compressed face manipulation (FF++ ROC-AUC = 0.5913).
Comments7 pages, 3 figures, 7 tables
