Quantum-Inspired Fine-Tuning for Few-Shot AIGC Detection via Phase-Structured Reparameterization
Authors: Kaiyang Xing, Han Fang, Zhaoyun Chen, Zhonghui Li, Yang Yang, Weiming Zhang, Guoping Guo
Organizations: Anhui University, Hefei, China · National University of Singapore, Singapore, Singapore · Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, China · University of Science and Technology of China, Hefei, China
Recent studies show that quantum neural networks (QNNs) generalize well in few-shot regimes. To extend this advantage to large-scale tasks, we propose Q-LoRA, a quantum-enhanced fine-tuning scheme that integrates lightweight QNNs into the low-rank adaptation (LoRA) adapter. Applied to AI-generated content (AIGC) detection, Q-LoRA consistently outperforms standard LoRA under few-shot settings. We analyze the source of this improvement and identify two possible structural inductive biases from QNNs: (i) phase-aware representations, which encode richer information across orthogonal amplitude-phase components, and (ii) norm-constrained transformations, which stabilize optimization via inherent orthogonality. However, Q-LoRA incurs non-trivial overhead due to quantum simulation. Motivated by our analysis, we further introduce H-LoRA, a fully classical variant that applies the Hilbert transform within the LoRA adapter to retain similar phase structure and constraints. Experiments on few-shot AIGC detection show that both Q-LoRA and H-LoRA outperform standard LoRA by over 5% accuracy, with H-LoRA achieving comparable accuracy at significantly lower cost in this task.
Figures & tables
Figure 1: Method of LoRA and Q-LoRA applied to the CLIP model, and their detection accuracy for AI-generated image detection tasks under different training samples.
Model SAMPLES
Metrics
CLIP
LoRA
Q-LoRA
H-LoRA
200
400
800
200
400
800
200
400
800
200
400
800
Midjourney
AUC(%)
98.38
98.29
98.38
94.02
95.78
97.59
98.62
98.74
98.68
96.31
97.80
97.91
ACC(%)
93.40
92.85
91.85
82.60
80.55
78.70
90.40
86.15
81.75
86.10
81.90
84.85
PR(%)
93.86
90.96
88.56
77.80
73.42
71.37
87.62
79.93
73.45
80.37
75.08
77.95
RE(%)
93.00
95.50
96.50
93.60
97.40
98.20
96.00
98.60
99.50
96.40
98.40
98.70
Wukong
AUC(%)
97.31
99.00
99.27
97.22
99.34
99.85
99.43
99.83
99.97
99.18
99.84
99.94
Table 1: Performance comparison under different few-shot settings for AI-generated image detection. For each metric, the best and second-best results within the same training sample size are highlighted in bold and underline, respectively.
Figure 3: Q-LoRA and H-LoRA t-SNE visualization
Figure 4: Attention visualization of Q-LoRA and H-LoRA
Method
Samples
AUC(%)
F1(%)
ACC(%)
EER
50
90.85
72.54
72.99
0.1616
Whisper
100
93.71
87.42
88.19
0.1260
200
95.82
89.93
90.69
0.0957
50
92.86
78.89
80.34
0.1358
LoRA
100
97.45
89.39
90.40
0.0711
200
99.04
95.54
95.67
0.0431
Table 2: Algorithm performance with varying sample sizes in audio forgery detection
Method
H-LoRA
Q-LoRA
Inference Time(sec)
0.09
65.68
Training Time(epoch/sec)
4.07
2088.34
Additional Parameter
0
24
Table 3: Efficiency comparison between H-LoRA and Q-LoRA.
Figure 5: Accuracy of LoRA, Q-LoRA and H-LoRA under different ranks and training sample sizes
Method
H-LoRA
Linear
Tanh
Sigmoid
Silu
AUC(%)
97.12
96.32
96.51
96.89
97.01
ACC(%)
89.94
87.79
88.30
87.19
88.36
Pr(%)
87.03
85.57
85.70
84.20
85.91
Re(%)
96.40
96.44
95.14
95.87
95.59
Table 4: Performance of different types of intermediate layers in LoRA
Method
H-LoRA
1
2
3
4
5
AUC(%)
97.12
96.32
97.39
97.52
97.58
97.61
ACC(%)
89.94
87.79
88.90
89.60
89.10
89.30
Pr(%)
87.03
85.57
86.04
86.25
85.77
86.68
Re(%)
96.40
96.44
96.41
97.19
97.17
96.23
Table 5: Performance of Number of Intermediate Linear Layers in LoRA
Variational quantum algorithms often encounter barren plateaus, where cost gradients decay rapidly with increasing circuit depth, undermining the trainability of parameterized quantum circuits. This paper evaluates AdaInit (Adaptive Initialization), proposed by Zhuang and Cunningham, which uses large language models to propose initial parameters for quantum neural networks. We study a simplified single-query AdaInit variant paired with GPU-accelerated simulation in NVIDIA CUDA-Q and apply it to binary classification on the DMR-IR mammography dataset. AdaInit delivers 14.6 times higher gradient variance at initialization than random initialization (0.0095 vs. 0.0006), producing 160 times faster convergence (1.1s vs. 176 s) while maintaining the same classification accuracy of 61.4 percent. We provide theoretical analysis grounded in the geometry of parameterized circuit landscapes and show empirically that LLM-guided initialization places the optimizer in trainable regions of parameter space. Beyond performance, our results indicate that a single LLM query can yield informative parameters without iterative refinement, suggesting a low-overhead path to improved trainability. The findings validate AdaInit in a medical imaging setting and demonstrate its compatibility with GPU-accelerated quantum backends for practical speedups.
Riza Alaudin Syah, Irwan Alnarus Kautsar, Haza Nuzly Bin Abdull Hamed
Faculty of Computing Universiti Teknologi Malaysia Johor Bahru, Malaysia · Remote Student based in Jakarta, Indonesia · Informatics Department Faculty of Science and Technology Universitas Muhammadiyah Sidoarjo Sidoarjo, Indonesia
Quantum computing provides a powerful paradigm for representing and transforming high-dimensional information through superposition, entanglement, and measurement-induced nonlinear features. While current quantum hardware is not yet practical for direct large-scale vision-language model (VLM) inference, simulated quantum computation can be used during model construction to generate structured parameters for compact classical AI systems. We build RiverONE, a lightweight vision-language model for quantum calibration plot understanding, using simulated quantum computation. It employs a specialized visual encoder and an InternVL-based language backbone. To compensate for compression-induced information loss, we introduce quantum-generated parameters, which are materialized as classical tensors after training. This allows RiverONE to run entirely on classical GPUs at inference time, with no quantum hardware or runtime quantum simulation. With approximately 1.9 billion parameters, RiverONE achieves at least 95% of the performance of NVIDIA Ising Calibration 1 on quantum calibration plot understanding tasks while using less than 10% of its parameter count. These results suggest that simulated quantum computation can serve as a practical construction-stage mechanism for building lightweight, knowledge-intensive scientific VLMs. Our code is available at https://github.com/THeWakeSystems/RiverOne.
Quantized fine-tuning (QLoRA) saves memory but not time. It dequantizes every 4-bit weight on the fly, so it trains more slowly than fp16 LoRA. We present AQLoRA (Adaptive-Quantization LoRA), a recipe that buys part of that time back. One CPU pass ranks layers by NF4 reconstruction error, with no search and no calibration data, and keeps a fixed fraction of them in fp16. Those layers skip dequantization, which is where the speed comes from. A quality setting adapts every layer. A speed setting adapts only the top blocks, so the backward pass stops early. The selection matches the accuracy of Unsloth's hand-curated dynamic-4bit list. We evaluate on Commonsense-170K across six models and four architecture families, from 1.4B to 14B. The speed setting trains 11.3 +/- 2.3% faster than well-tuned QLoRA and gives up 0.1 to 1.5 accuracy points. It was faster in all thirteen independent timing sessions, at worst by 7%. The quality setting trains 4.8 +/- 2.0% faster, matches QLoRA's accuracy on every model, and is within a point of fp16 LoRA. The memory premium depends on the architecture: 0.06 to 0.8 GiB on the 1-4B models, 0.5 GiB at 7B, 1.9 GiB at 14B. These error bars are measured between independent sessions, not within one. Timing on shared hardware needs three rules: fix the measurement duration, not the step count; measure the noise floor from a duplicated arm; and repeat whole sessions, since on one model three sessions that each passed the floor test spread over 8.2 points. We validate the recipe with controls and report the two that failed. Choosing adapter layers by weight density is no better than random. Choosing protected layers by quantization error is not either, on five models at three budgets with memory-matched controls and downstream on two. The number of protected layers sets the speed; which layers they are does not measurably change accuracy.