Deep Learning of Potential Outcomes

Koch, Bernard; Sainburg, Tim; Geraldo, Pablo; Jiang, Song; Sun, Yizhou; Foster, Jacob G.

doi:10.48550/arxiv.2110.04442

Cited by 6 publications

(10 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This review work complements two excellent and extensive overviews of causal inference methods by Yao et al [4] and Koch et al [21]. The former focused on categorizing existing causal inference methods.…”

Section: Related Workmentioning

confidence: 85%

Review of Deep Learning Methods for Individual Treatment Effect Estimation with Automatic Hyperparameter Optimization

Buchwald¹,

Sirazitdinov²,

Hesser³

et al. 2022

Preprint

View full text Add to dashboard Cite

<p>Abstract—Estimation of individual treatment effect (ITE) for</p> <p>different types of treatment is a common challenge in therapy assessments, clinical trials and diagnosis. Deep learning methods, namely representation based, adversarial, and variational, have shown promising potential in ITE estimation. However, it was unclear whether the hyperparameters of the originally proposed methods were well optimized for different benchmark datasets. To solve these problems, we created a public code library containing representation-based, adversarial, and variational methods written in TensorFlow. In order to have a broader collection of ITE estimation methods, we have also included neural network based meta-learners. The code library is made accessible for reproducibility and facilitating future works in the field of causal inference. Our results demonstrate that performance of most methods can be improved using automatic hyperparameter optimization. Additionally, we review the methods and compare the performance of the optimized models from our library on publicly available datasets. The potential of hyperparameter optimization may encourage researchers to focus on this aspect when creating new methods for inferring individual treatment effect.</p>

show abstract

Section: Related Workmentioning

confidence: 85%

Review of Deep Learning Methods for Individual Treatment Effect Estimation with Automatic Hyperparameter Optimization

Buchwald¹,

Sirazitdinov²,

Hesser³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The simulation we have set up for assessing predictive performance shows promise but also reveals limitations that are discussed further below. However, in a stringent machine learning perspective, one might also consider adopting deep neural networks as causal estimators, a novel and vibrant research area [ 97 , 98 ]. Preliminary studies in this field claim, for low bias, suitability for estimating the heterogeneous treatment effects and providing opportunities to predict causal effects in untreated populations beyond the original sample.…”

Section: Discussionmentioning

confidence: 99%

“…Preliminary studies in this field claim, for low bias, suitability for estimating the heterogeneous treatment effects and providing opportunities to predict causal effects in untreated populations beyond the original sample. However, there is still little consensus on important practical considerations needed to deploy these tools in the wild [ 97 ]. For what concerns causal inference for affective/social behaviour, the long-term most promising avenue is offered by deep learning of the causal structure of dynamic systems and time series data [ 99 ].…”

Section: Discussionmentioning

confidence: 99%

Inferring Causal Factors of Core Affect Dynamics on Social Participation through the Lens of the Observer

D’Amelio

Patania

Buršić

et al. 2023

Sensors

View full text Add to dashboard Cite

A core endeavour in current affective computing and social signal processing research is the construction of datasets embedding suitable ground truths to foster machine learning methods. This practice brings up hitherto overlooked intricacies. In this paper, we consider causal factors potentially arising when human raters evaluate the affect fluctuations of subjects involved in dyadic interactions and subsequently categorise them in terms of social participation traits. To gauge such factors, we propose an emulator as a statistical approximation of the human rater, and we first discuss the motivations and the rationale behind the approach.The emulator is laid down in the next section as a phenomenological model where the core affect stochastic dynamics as perceived by the rater are captured through an Ornstein–Uhlenbeck process; its parameters are then exploited to infer potential causal effects in the attribution of social traits. Following that, by resorting to a publicly available dataset, the adequacy of the model is evaluated in terms of both human raters’ emulation and machine learning predictive capabilities. We then present the results, which are followed by a general discussion concerning findings and their implications, together with advantages and potential applications of the approach.

show abstract

“…Causal prediction through DL algorithms has been recently drawing much attention. Koch et al (2021) reviewed four different approaches to deep causal estimation including (i) meta-learners (Künzel et al, 2019) like S-learners, T-leaners, and X-learners, (ii) balancing through representation learning such as TARNet (Shalit et al, 2017) and CFRNet (Johansson et al, 2018(Johansson et al, , 2020, (iii) extension with inverse propensity score weighting such as targeted maximum likelihood estimation (TMLE, Van der Laan et al (2011)) and Dragonnet and targeted regularization (Shi et al, 2019), and (iv) adversarial training of generative models such as GANITE (Yoon et al, 2018). In particular, TARNet utilizes the loss functions that minimize the mean square errors (MSE) of the observed vs. predicted causal outcomes after representation learning, i.e., after deconfounding the treatment from the outcome by forcing the treated and control covariate distributions to get closer together, and CFRNet additionally minimizes the distance between treatment indicator and covariate distributions.…”

Section: Related Work and Contributionmentioning

confidence: 99%

“…To calculate the ATE, the TMLE is a two-step procedure, the first step being an input-output modeling without adjusting for the inverse propensity score weights and the second step being the estimation step for the fluctuation parameter i as a constant regression coefficient for the slope between the inverse propensity score weights and the observed outcome.It was noted (Koch et al, 2021) that this would be equivalent to minimizing the weighted term involving i in equation ( 5).…”

Section: No Censoring (Nc) Casesmentioning

confidence: 99%

CausalDeepCENT: Deep Learning for Causal Prediction of Individual Event Times

Jeong¹,

Jia²

2022

Preprint

View full text Add to dashboard Cite

Deep learning (DL) has recently drawn much attention in image analysis, natural language process, and high-dimensional medical data analysis. Under the causal direct acyclic graph (DAG) interpretation, the input variables without incoming edges from parent nodes in the DL architecture maybe assumed to be randomized and independent of each other. As in a regression setting, including the input variables in the DL algorithm would reduce the bias from the potential confounders. However, failing to include a potential latent causal structure among the input variables affecting both treatment assignment and the output variable could be additional significant source of bias. The primary goal of this study is to develop new DL algorithms to estimate causal individual event times for time-to-event data, equivalently to estimate the causal time-to-event distribution with or without right censoring, accounting for the potential latent structure among the input variables. Once the causal individual event times are estimated, it would be straightforward to estimate the causal average treatment effects as the differences in the averages of the estimated causal individual event times. A connection is made between the proposed method and the targeted maximum likelihood estimation (TMLE). Simulation studies are performed to assess improvement in prediction abilities of the proposed methods by using the mean square error (MSE)-based method and rank-based C-Index metric. The simulation results indicate that improvement on the prediction accuracy could be substantial particularly when there is a collider among the input variables. The proposed method is illustrated with a publicly available and influential breast cancer data set. The proposed method has been implemented by using PyTorch and uploaded at https://github.com/yicjia/CausalDeepCENT.

show abstract

Deep Learning of Potential Outcomes

Cited by 6 publications

References 22 publications

Review of Deep Learning Methods for Individual Treatment Effect Estimation with Automatic Hyperparameter Optimization

Review of Deep Learning Methods for Individual Treatment Effect Estimation with Automatic Hyperparameter Optimization

Inferring Causal Factors of Core Affect Dynamics on Social Participation through the Lens of the Observer

CausalDeepCENT: Deep Learning for Causal Prediction of Individual Event Times

Contact Info

Product

Resources

About