Estimating Visual Attribute Effects in Advertising from Observational Data: A Deepfake-Informed Double Machine Learning Approach

Yizhi Liu; Balaji Padmanabhan; Siva Viswanathan

cs.AIMar 2, 2026

Estimating Visual Attribute Effects in Advertising from Observational Data: A Deepfake-Informed Double Machine Learning Approach

Authors: Yizhi Liu, Balaji Padmanabhan, Siva Viswanathan

Organizations: Robert H. Smith School of Business, University of Maryland, College Park, MD, USA

Abstract

Digital advertising increasingly relies on visual content, yet marketers lack rigorous methods for understanding how specific visual attributes causally affect consumer engagement. This paper addresses a fundamental methodological challenge: estimating causal effects when the treatment, such as a model's skin tone, is an attribute embedded within the image itself. Standard approaches like Double Machine Learning (DML) fail in this setting because vision encoders entangle treatment information with confounding variables, producing severely biased estimates. We develop DICE-DML (Deepfake-Informed Control Encoder for Double Machine Learning), a framework that leverages generative AI to disentangle treatment from confounders. The approach combines three mechanisms: (1) deepfake-generated image pairs that isolate treatment variation; (2) DICE-Diff adversarial learning on paired difference vectors, where background signals cancel to reveal pure treatment fingerprints; and (3) orthogonal projection that geometrically removes treatment-axis components. In simulations with known ground truth, DICE-DML reduces root mean squared error by 73-97% compared to standard DML, with the strongest improvement (97.5%) at the null effect point, demonstrating robust Type I error control. Applying DICE-DML to 232,089 Instagram influencer posts, we estimate the causal effect of skin tone on engagement. Standard DML produces diagnostically invalid results (negative outcome R^2), while DICE-DML achieves valid confounding control (R^2 = 0.63) and estimates a marginally significant negative effect of darker skin tone (-522 likes; p = 0.062), substantially smaller than the biased standard estimate. Our framework provides a principled approach for causal inference with visual data when treatments and confounders coexist within images.

Figures & tables

Symbol	Description
$T\in\{0,1\}$	Binary treatment (e.g., skin tone)
$Y\in\mathbb{R}$	Outcome (e.g., number of likes)
$I$	Original image
$I^{\prime}$	Deepfake counterpart of $I$
$Z=f(I)$	Learned representation of image $I$
$X^{*}$	True (unobserved) confounders

Table 1. Summary of Key Notation

Figure 1. The DICE-DML Framework. The pipeline consists of five stages: (1) Input pairs of original images and their deepfake counterparts with altered treatment attribute; (2) A shared encoder (e.g., ResNet+MLP) extracts representations $Z=f(I)$ and $Z^{\prime}=f(I^{\prime})$ ; (3) The DICE-Diff adversarial learning operates on difference vectors $\Delta=Z-Z^{\prime}$ where background signals cancel, training discriminators to detect treatment while the encoder learns to fool them via gradient reversal; (4) Orthogonal projection computes sample-wise treatment directions from paired differences and projects them out to obtain $Z_{\text{clean}}$ ; (5) The cleaned representations feed into standard DML with outcome and propensity models for treatment effect estimation. A diagram showing the DICE-DML framework with five stages.

Method	Avg $\|\text{Bias}\|$	Avg RMSE	Reduction
Naive OLS	0.067	0.168	—
DML (Original)	0.115	0.317	(baseline)
DML (DICE)	0.051	0.052	73–97%

Table 2. Simulation Results with Known Ground-Truth Effects

Figure 2. Asymptotic Normality of Treatment Effect Estimators. Distribution of standardized t-statistics at $\tau=0$ across 200 simulations. Naive estimation without controls (blue) produces highly dispersed estimates (std = 11.53). Standard DML with original ResNet features (orange) remains overdispersed (std = 2.83), indicating treatment leakage distorts variance. DICE-DML (green) closely tracks the theoretical $N(0,1)$ distribution (std $\approx$ 0.98), confirming valid inference and proper Type I error control. Histogram showing three distributions of t-statistics, with DICE-DML matching standard normal.

Method	$\tau$ (likes)	95% CI	$Y$ $R^{2}$	$T$ $R^{2}$
Naive OLS	$+74$	—	—	—
DML (Original)	$-1{,}455^{**}$	$[-2494,-415]$	$-0.003$	0.251
DML (DICE)	$\mathbf{-522^{\dagger}}$	$\mathbf{[-1070,+26]}$	$\mathbf{0.626}$	$\mathbf{0.149}$

Table 3. Causal Effect Estimates for Skin Tone on Engagement

Figure 3. Distribution of difference vector norms between original and deepfake image pairs. DICE-DML reduces the mean L2 distance from 3.6 to 0.9 (75% reduction), indicating that the encoder learns to produce similar representations for image pairs that differ only in treatment (skin tone), while preserving other visual information. This compression does not reflect representation collapse, as the same embeddings achieve $Y$ $R^{2}=0.63$ for outcome prediction. Histogram comparing difference vector norm distributions for Original ResNet and DICE embeddings.

Specification	$\tau$ (std)	SE	$\tau$ (likes)	$Y$ $R^{2}$
RF (shallow)	$-0.0106$	0.0097	$-358$	0.716
RF (deep)	$-0.0116$	0.0097	$-391$	0.720
GBM	$-0.0111$	0.0100	$-375$	0.702
Ridge	$-0.0103$	0.0154	$-348$	0.341

Table 4. Robustness to Nuisance Model Specification

Configuration	RMSE	$Y$ $R^{2}$	$T$ $R^{2}$	$\\|\Delta\\|$
Baseline (encoder only)	0.050	0.28	0.25	3.57
Adversarial only	0.042	0.90	0.14	3.57
Projection only	0.051	$-$ 0.07	0.22	0.18
Full DICE-DML	0.043	0.63	0.15	0.87

Table 5. Ablation Study: Component Contributions

Causal Effect Estimation

Causal effect estimation measures how outcomes would change under interventions, using experiments or observational data and explicit assumptions.

See all 264 papers

Oct 16, 2026

Causal Value Attribution of Marketing Labels in LLM-Generated Ads: A Deconfounder Approach

As AI-generated content becomes commoditized, economic value increasingly shifts from final artifacts to the upstream expertise used to specify inputs that steer LLM generation. Yet organizations lack methods for valuing this expertise and identifying which individual skills or configurations causally improve performance. This study examines generative marketing, where structured expert labels serve as skill primitives conditioning LLM-generated advertisements. Estimating their causal effects is difficult because outcomes reflect stochastic LLM mediation, high-dimensional label interactions, and multimodal confounding. This study proposes D-VAE (Deconfounder Variational Autoencoder), a counterfactual-inference framework integrating a multi-treatment deconfounder with causal representation learning. D-VAE recovers causally disentangled expert-labeling logic and derives PAC-based bounds on omitted-variable bias from single-treatment confounders. Applied to 29,177 foundation-makeup ad creatives, D-VAE outperforms state-of-the-art benchmarks. A field experiment shows that recommended configurations increase engagement by 32.7% and perform statistically indistinguishably from human experts, while a simulation study confirms causal attribution under complex confounding and treatment interactions. Overall, the study shifts value assessment from AI-generated products to skills, providing a foundation for pricing and optimizing AI-augmented skill portfolios.

Danhong You, Wei Jiang, Zhiqiang +1

Eric · Shanghai Jiao Tong University · University of Texas at Dallas

Causal Inference Causal Inference with Unobserved Confounding Online Advertising Causal Representation Learning

Ask PDF

May 24, 2026

Disentangled Double Machine Learning for Accurate Causal Effect Estimation

Confounding bias is a key challenge in causal effect estimation from observational data. Double Machine Learning (DML) addresses this issue by estimating treatment and outcome nuisance functions, constructing treatment and outcome residuals, and estimating causal effects from the residuals. However, DML often produces biased and unstable estimates in highdimensional or finite-sample scenarios. One reason is that DML estimates nuisance functions using all covariates without disentangling distinct latent factors, resulting in unreliable nuisance function estimation. Another is that imprecise nuisance estimation further introduces residual dependence between the treatment residual and the remaining outcome error, undermining the accuracy of causal effect estimates. To address these issues, in this paper, we propose Disentangled Double Machine Learning (DDML), a novel algorithm that integrates two key strategies. First, a causal role disentanglement strategy decomposes covariates into confounders, treatment-specific factors, and outcomespecific factors for enabling reliable nuisance function estimation. And second, a residual dependence orthogonalization strategy mitigates residual dependence caused by nuisance estimation errors for enhancing the precision of causal effect estimates. Experimental results on synthetic, semi-synthetic, and real-world datasets demonstrate that DDML significantly outperforms 13 state-of-the-art baseline algorithms in both MAE and RMSE.

Guodu Xiang, Kui Yu, Yujie Wang +3

School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230601, China · School of Computer and Information Technology, Shanxi University, Taiyuan 030006, China

Double ML Causal Effect Estimation

Ask PDF

Oct 15, 2026

Deep Learning for Deepfake Image Detection: Comparative Model Evaluation and Gradient-Based Visual Attribution

Synthetic facial imagery has made the distinction between authentic and generated content increasingly important for digital forensics and online trust. This study examines an explainable deep-learning workflow for image-level deepfake detection using two contrasting classification strategies: a compact convolutional model trained from scratch and an ImageNet-pretrained MobileNetV2 model adapted through two-stage transfer learning. Experiments were conducted with the balanced 140k Real and Fake Faces dataset, comprising 140,000 facial images. Under the recorded validation protocol, the augmented from-scratch model reached 84.15% accuracy (loss 0.3625), compared with 95.90% accuracy (loss 0.1074) for the fine-tuned pretrained model. Gradient-weighted class activation mapping (Grad-CAM) was used to visualize image regions associated with predictions, and a Flask-based prototype was developed for side-by-side inspection. These observations support the practical value of pretrained visual representations in the studied setting while underscoring the need to interpret validation performance alongside dataset provenance, explanation limitations, and evaluation scope. The reported results are validation results, not held-out test estimates.

Saad Korchi, Abderrahim Waga

Euromed University of Fez, Fez, Morocco

Deepfake Detection Explainable Deepfake Detection Transfer Learning

Ask PDF

Abstract

Causal Effect Estimation

Explore similar work

Causal Value Attribution of Marketing Labels in LLM-Generated Ads: A Deconfounder Approach

Disentangled Double Machine Learning for Accurate Causal Effect Estimation

Deep Learning for Deepfake Image Detection: Comparative Model Evaluation and Gradient-Based Visual Attribution