Latent Regularization in Generative Test Input Generation

Giorgi Merabishvili; Oliver Weißl; Andrea Stocco

cs.SEFeb 17, 2026

Latent Regularization in Generative Test Input Generation

Authors: Giorgi Merabishvili, Oliver Weißl, Andrea Stocco

Organizations: North Carolina State University USA · Technical University of Munich & fortiss GmbH Germany

Abstract

This study investigates the impact of regularization of latent spaces through truncation on the quality of generated test inputs for deep learning classifiers. We evaluate this effect using style-based GANs, a state-of-the-art generative approach, and assess quality along three dimensions: validity, diversity, and fault detection. We evaluate our approach on the boundary testing of deep learning image classifiers across three datasets, MNIST, Fashion MNIST, and CIFAR-10. We compare two truncation strategies: latent code mixing with binary search optimization and random latent truncation for generative exploration. Our experiments show that the latent code-mixing approach yields a higher fault detection rate than random truncation, while also improving both diversity and validity.

Figures & tables

$\psi{=}1.0$	$0.8$	$0.6$	$0.4$	$0.2$

Figure 1. Truncation on coarse layers. A fixed latent seed at

\psi_{T}=1.0

and progressively lower

\psi_{T}

values increase fidelity and reduce diversity.

Figure 2. Test input generation with truncation workflow.

Panel of Figure 3 — Figure 3. Truncation-only search. Truncation refines a human-invalid baseline into a human-valid image and flips the SUT prediction.

Figure 4. Minimal truncation

\psi^{\star}

in MNIST (

n=25

Panel of Figure 5 — Figure 5. Target revelation via truncation. Lowering $\psi$ causes human-valid seed to flip under truncation alone.

Dataset	Technique	Setting	Seeds	Human-val	Human-Val. Rate	LPIPS	Fault Rate
MNIST	Style-Mixing	No trunc.	504	11	44%	0.194	2.18%
		$\psi{=}0.9$	518	12	48%	0.224	2.31%
		$\psi{=}0.8$	461	16	64%	0.215	3.47%
		$\psi{=}0.7$	549	14	56%	0.190	2.55%
		$\psi{=}0.6$	455	20	80%	0.152	4.39%
		$\psi{=}0.5$	448	17	68%	0.143	3.79%

Table 1. Results across datasets, truncation levels, and probing techniques. Each entry reports the number of sampled seeds, human-validated inputs, validation rate, mean LPIPS , fault detection rate, and efficiency (seeds per validated input).

Figure 6. MNIST: adaptive truncation, human-validated, fault-revealing test cases.

Figure 7. Cases from MNIST where the first flip occurs at $\psi^{\star}$ .

Panel of Figure 8 — Figure 8. Layer-specific truncation sweeps (same seed). Columns show decreasing $\psi$ from 1.0 to 0.1; rows indicate which layers are truncated: up to layer 2 (coarse), up to layer 5 (coarse+middle), and up to layer 8 (all layers).

Automated Test Generation

Automated test generation creates inputs, scenarios, and checks to expose software and system defects with less manual effort.

See all 132 papers

Latent Anchor-Driven Test Generation for Deep Neural Networks
Jun 3, 2026Bin Duan, Matthew B. Dwyer, Guowei YangDNN Testing Automated Test Generation
School of Electrical Engineering and Computer Science, The University of Queensland, Australia · Department of Computer Science, University of Virginia, United States
HyperNet-Adaptation for Diffusion-Based Test Case Generation
Jan 21, 2026Oliver Weißl, Vincenzo Riccio, Severin Kacianka +1Hypernetworks Automated Test Generation
Technical University of Munich, Germany · University of Udine, Italy · Independent Researcher, Germany +1
Quantifying the Effect of Test Set Contamination on Generative Evaluations
Jan 7, 2026Rylan Schaeffer, Joshua Kazdan, Baber Abbasi +8Language Model Generation Evaluation Benchmark Contamination
Stanford Computer Science · Stanford Statistics · EleutherAI +3

Jun 3, 2026J/K move · Enter open · S save

Latent Anchor-Driven Test Generation for Deep Neural Networks

Bin Duan, Matthew B. Dwyer, Guowei Yang

School of Electrical Engineering and Computer Science, The University of Queensland, Australia · Department of Computer Science, University of Virginia, United States

DNN Testing Automated Test Generation

Open paper AskPDF arXiv

Deep Neural Networks (DNNs) are increasingly being deployed in security-critical and safety-sensitive applications, which makes rigorous testing essential to identify and mitigate model weaknesses. Existing DNN testing approaches explore either the input space or a learned latent space. While latent-space generation can better maintain plausibility than direct input-space mutation, current methods still face a trade-off among exploration controllability, failure diversity, and seed-relative semantic drift. To overcome these limitations, we propose Latte, a black-box testing framework that generates semantically proximate, diverse, and fault-revealing test cases by leveraging the latent space. Specifically, Latte encodes each input seed with a pre-trained VQ-VAE and performs a seed-centered, one-step latent mutation along directions defined by anchors sampled from alternative classes, followed by quantization and decoding back to the input space. This explores local neighborhoods around each seed within the learned latent manifold, resulting in a larger number and broader diversity of oracle-triggering prediction discrepancies under the same budget. We evaluated Latte on 5 datasets and 10 DNN models in single-model and multi-model testing scenarios. Across the evaluated datasets and models, Latte improves fault exposure and behavioral diversity under matched testing budgets. Under the single-model setting, it also maintains low seed-relative semantic drift with respect to the source seeds.

Abstract

Automated Test Generation

Explore similar work

Latent Anchor-Driven Test Generation for Deep Neural Networks

HyperNet-Adaptation for Diffusion-Based Test Case Generation

Quantifying the Effect of Test Set Contamination on Generative Evaluations