Showing papers on "Counterfactual conditional published in 2022"

Model agnostic generation of counterfactual explanations for molecules

Abstract: An outstanding challenge in deep learning in chemistry is its lack of interpretability. The inability of explaining why a neural network makes a prediction is a major barrier to deployment of AI models. This not only dissuades chemists from using deep learning predictions, but also has led to neural networks learning spurious correlations that are difficult to notice. Counterfactuals are a category of explanations that provide a rationale behind a model prediction with satisfying properties like providing chemical structure insights. Yet, counterfactuals have been previously limited to specific model architectures or required reinforcement learning as a separate process. In this work, we show a universal model-agnostic approach that can explain any black-box model prediction. We demonstrate this method on random forest models, sequence models, and graph neural networks in both classification and regression.

Learning Fair Node Representations with Graph Counterfactual Fairness

Abstract: Fair machine learning aims to mitigate the biases of model predictions against certain subpopulations regarding sensitive attributes such as race and gender. Among the many existing fairness notions, counterfactual fairness measures the model fairness from a causal perspective by comparing the predictions of each individual from the original data and the counterfactuals. In counterfactuals, the sensitive attribute values of this individual had been modified. Recently, a few works extend counterfactual fairness to graph data, but most of them neglect the following facts that can lead to biases: 1) the sensitive attributes of each node's neighbors may causally affect the prediction w.r.t. this node; 2) the sensitive attributes may causally affect other features and the graph structure. To tackle these issues, in this paper, we propose a novel fairness notion - graph counterfactual fairness, which considers the biases led by the above facts. To learn node representations towards graph counterfactual fairness, we propose a novel framework based on counterfactual data augmentation. In this framework, we generate counterfactuals corresponding to perturbations on each node's and their neighbors' sensitive attributes. Then we enforce fairness by minimizing the discrepancy between the representations learned from the original graph and the counterfactuals for each node. Experiments on both synthetic and real-world graphs show that our framework outperforms the state-of-the-art baselines in graph counterfactual fairness, and also achieves comparable prediction performance.

Counterfactual explanations and how to find them: literature review and benchmarking

Abstract: Abstract Interpretable machine learning aims at unveiling the reasons behind predictions returned by uninterpretable classifiers. One of the most valuable types of explanation consists of counterfactuals. A counterfactual explanation reveals what should have been different in an instance to observe a diverse outcome. For instance, a bank customer asks for a loan that is rejected. The counterfactual explanation consists of what should have been different for the customer in order to have the loan accepted. Recently, there has been an explosion of proposals for counterfactual explainers. The aim of this work is to survey the most recent explainers returning counterfactual explanations. We categorize explainers based on the approach adopted to return the counterfactuals, and we label them according to characteristics of the method and properties of the counterfactuals returned. In addition, we visually compare the explanations, and we report quantitative benchmarking assessing minimality, actionability, stability, diversity, discriminative power, and running time. The results make evident that the current state of the art does not provide a counterfactual explainer able to guarantee all these properties simultaneously.

A Practical Guide to Counterfactual Estimators for Causal Inference with Time‐Series Cross‐Sectional Data

Abstract: This paper introduces a simple framework of counterfactual estimation for causal inference with time-series cross-sectional data, in which we estimate the average treatment effect on the treated by directly imputing counterfactual outcomes for treated observations. We discuss several novel estimators under this framework, including the fixed effects counterfactual estimator, interactive fixed effects counterfactual estimator and matrix completion estimator. They provide more reliable causal estimates than conventional two-way fixed effects models when treatment effects are heterogeneous or unobserved time-varying confounders exist. Moreover, we propose a new dynamic treatment effects plot, along with several diagnostic tests, to help researchers gauge the validity of the identifying assumptions. We illustrate these methods with two political economy examples and develop an open-source package, fect, in both R and Stata to facilitate implementation.

Market counterfactuals and the specification of multiproduct demand: A nonparametric approach

Abstract: Demand estimates are essential for addressing a wide range of positive and normative questions in economics that are known to depend on the shape—and notably the curvature—of the true demand functions. The existing frontier approaches, while allowing flexible substitution patterns, typically require the researcher to commit to a parametric specification. An open question is whether these a priori restrictions are likely to significantly affect the results. To address this, I develop a nonparametric approach to estimation of demand for differentiated products, which I then apply to California supermarket data. While the approach subsumes workhorse models such as mixed logit, it allows consumer behaviors and preferences beyond standard discrete choice, including continuous choices, complementarities across goods, and consumer inattention. When considering a tax on one good, the nonparametric approach predicts a much lower pass‐through than a standard mixed logit model. However, when assessing the market power of a multiproduct firm relative to that of a single‐product firm, the models give similar results. I also illustrate how the nonparametric approach may be used to guide the choice among parametric specifications.

DoCoGen: Domain Counterfactual Generation for Low Resource Domain Adaptation

Abstract: Natural language processing (NLP) algorithms have become very successful, but they still struggle when applied to out-of-distribution examples. In this paper we propose a controllable generation approach in order to deal with this domain adaptation (DA) challenge. Given an input text example, our DoCoGen algorithm generates a domain-counterfactual textual example (D-con) - that is similar to the original in all aspects, including the task label, but its domain is changed to a desired one. Importantly, DoCoGen is trained using only unlabeled examples from multiple domains - no NLP task labels or parallel pairs of textual examples and their domain-counterfactuals are required. We show that DoCoGen can generate coherent counterfactuals consisting of multiple sentences. We use the D-cons generated by DoCoGen to augment a sentiment classifier and a multi-label intent classifier in 20 and 78 DA setups, respectively, where source-domain labeled data is scarce. Our model outperforms strong baselines and improves the accuracy of a state-of-the-art unsupervised DA algorithm.

C2L: Causally Contrastive Learning for Robust Text Classification

Abstract: Despite the super-human accuracy of recent deep models in NLP tasks, their robustness is reportedly limited due to their reliance on spurious patterns. We thus aim to leverage contrastive learning and counterfactual augmentation for robustness. For augmentation, existing work either requires humans to add counterfactuals to the dataset or machines to automatically matches near-counterfactuals already in the dataset. Unlike existing augmentation is affected by spurious correlations, ours, by synthesizing “a set” of counterfactuals, and making a collective decision on the distribution of predictions on this set, can robustly supervise the causality of each term. Our empirical results show that our approach, by collective decisions, is less sensitive to task model bias of attribution-based synthesis, and thus achieves significant improvements, in diverse dimensions: 1) counterfactual robustness, 2) cross-domain generalization, and 3) generalization from scarce data.

Conceptual challenges for interpretable machine learning

Abstract: Abstract As machine learning has gradually entered into ever more sectors of public and private life, there has been a growing demand for algorithmic explainability. How can we make the predictions of complex statistical models more intelligible to end users? A subdiscipline of computer science known as interpretable machine learning (IML) has emerged to address this urgent question. Numerous influential methods have been proposed, from local linear approximations to rule lists and counterfactuals. In this article, I highlight three conceptual challenges that are largely overlooked by authors in this area. I argue that the vast majority of IML algorithms are plagued by (1) ambiguity with respect to their true target; (2) a disregard for error rates and severe testing; and (3) an emphasis on product over process. Each point is developed at length, drawing on relevant debates in epistemology and philosophy of science. Examples and counterexamples from IML are considered, demonstrating how failure to acknowledge these problems can result in counterintuitive and potentially misleading explanations. Without greater care for the conceptual foundations of IML, future work in this area is doomed to repeat the same mistakes.

What would have happened? Counterfactuals, hypotheticals and causal judgements

Abstract: How do people make causal judgements? In this paper, I show that counterfactual simulations are necessary for explaining causal judgements about events, and that hypotheticals do not suffice. In two experiments, participants viewed video clips of dynamic interactions between billiard balls. In Experiment 1, participants either made hypothetical judgements about whether ball B would go through the gate if ball A were not present in the scene, or counterfactual judgements about whether ball B would have gone through the gate if ball A had not been present. Because the clips featured a block in front of the gate that sometimes moved and sometimes stayed put, hypothetical and counterfactual judgements came apart. A computational model that evaluates hypotheticals and counterfactuals by running noisy physical simulations accurately captured participants’ judgements. In Experiment 2, participants judged whether ball A caused ball B to go through the gate. The results showed a tight fit between counterfactual and causal judgements, whereas hypotheticals did not predict causal judgements. I discuss the implications of this work for theories of causality, and for studying the development of counterfactual thinking in children. This article is part of the theme issue ‘Thinking about possibilities: mechanisms, ontogeny, functions and phylogeny’.

Keep Your Friends Close and Your Counterfactuals Closer: Improved Learning From Closest Rather Than Plausible Counterfactual Explanations in an Abstract Setting

Abstract: Counterfactual explanations (CFEs) highlight changes to a model’s input that alter its prediction in a particular way. s have gained considerable traction as a psychologically grounded solution for explainable artificial intelligence (XAI). Recent innovations introduce the notion of plausibility for automatically generated s, enhancing their robustness by exclusively creating plausible explanations. However, practical benefits of this constraint on user experience are yet unclear. In this study, we evaluate objective and subjective usability of plausible s in an iterative learning task. We rely on a game-like experimental design, revolving around an abstract scenario. Our results show that novice users benefit less from receiving plausible rather than closest s that induce minimal changes leading to the desired outcome. Responses in a post-game survey reveal no differences for subjective usability between both groups. Following the view of psychological plausibility as comparative similarity, users in the closest condition may experience their s as more psychologically plausible than the computationally plausible counterpart. In sum, our work highlights a little-considered divergence of definitions of computational plausibility and psychological plausibility, critically confirming the need to incorporate human behavior, preferences and mental models already at the design stages of XAI. All source code and data of the current study are available: https://github.com/ukuhl/PlausibleAlienZoo

What Can Time-Series Regressions Tell Us About Policy Counterfactuals?

Abstract: We show that, in a general family of linearized structural macroeconomic models, knowledge of the empirically estimable causal effects of contemporaneous and news shocks to the prevailing policy rule is sufficient to construct counterfactuals under alternative policy rules. If the researcher is willing to postulate a loss function, our results furthermore allow her to recover an optimal policy rule for that loss. Under our assumptions, the derived counterfactuals and optimal policies are robust to the Lucas critique. We then discuss strategies for applying these insights when only a limited amount of empirical causal evidence on policy shock transmission is available.

Patterned Outcomes, Unpatterned Counterfactuals, and Spurious Results: Perinatal Health Outcomes Following COVID-19

Abstract: Abstract The epidemiologic literature estimating the indirect or secondary effects of the COVID-19 pandemic on pregnant people and gestation continues to grow. Our assessment of this scholarship, however, leads us to suspect that the methods most commonly used may lead researchers to spurious inferences. This suspicion arises because the methods do not account for temporal patterning in perinatal outcomes when deriving counterfactuals, or estimates of the outcomes had the pandemic not occurred. We illustrate the problem in two ways. First, using monthly data from US birth certificates, we describe temporal patterning in five commonly used perinatal outcomes. Notably, for all but one outcome, temporal patterns appear more complex than much of the emerging literature assumes. Second, using data from France, we show that using counterfactuals that ignore this complexity produces spurious results. We recommend that subsequent investigations on COVID-19 and other perturbations use widely available time-series methods to derive counterfactuals that account for strong temporal patterning in perinatal outcomes.

Synthetic Control Method: A tool for comparative case studies in economic history

Abstract: The Synthetic Control Method (SCM) has become a widely used tool in both identifying and estimating the causal impact of policies, shocks, and interventions of interest on economic and social outcomes. The technique has become particularly popular in estimating the effect of these shocks on a single treated unit. As a transparent and data-driven statistical technique, the goal of the SCM is to construct an artificial control group for the treated unit that has similar pretreatment characteristics but has not undergone the treatment itself thus developing a plausible counterfactual against which impacts resulting from structural changes can be evaluated as part of a historical investigation. The method works well when the control group balances pre-intervention outcomes and auxiliary covariates as much as possible. In spite of its widespread adoption, the use of the SCM in comparative economic history has lagged behind other areas of economics. In this article, we critically review the properties of the SCM and discuss the necessary conditions for a plausible application of the technique to comparative economic history in support of research designed to answer some of the long-running historical questions and demonstrate the potential to use SCM in comparative economic history studies by estimating the impact of the oil discovery in the 1920s on Venezuela's long-term economic growth.

From Imitation to Innovation: Where Is All That Chinese R&D Going?

Abstract: We construct an endogenous growth model with random interactions where firms are subject to distortions. The TFP distribution evolves endogenously as firms seek to upgrade their technology over time either by innovating or by imitating other firms. We use the model to quantify the effects of misallocation on TFP growth in emerging economies. We structurally estimate the stationary state of the dynamic model targeting moments of the empirical distribution of R&D and TFP growth in China during the period 2007–2012. The estimated model fits the Chinese data well. We compare the estimates with those obtained using data for Taiwan and perform counterfactuals to study the effect of alternative policies. R&D misallocation has a large effect on TFP growth.

A Semistructural Methodology for Policy Counterfactuals

Abstract: I propose a methodology for constructing counterfactuals with respect to changes in policy rules that does not require fully specifying a particular model yet is not subject to Lucas critique. It applies to a class of dynamic stochastic models whose equilibria are well approximated by a linear representation. It rests on the insight that many such models satisfy a principle of counterfactual equivalence: they are observationally equivalent under a benchmark policy and yield an identical counterfactual equilibrium under an alternative one.

Evaluating and Mitigating Bias in Image Classifiers: A Causal Perspective Using Counterfactuals

Abstract: Counterfactual examples for an input—perturbations that change specific features but not others—have been shown to be useful for evaluating bias of machine learning models, e.g., against specific demographic groups. However, generating counterfactual examples for images is nontrivial due to the underlying causal structure on the various features of an image. To be meaningful, generated perturbations need to satisfy constraints implied by the causal model. We present a method for generating counterfactuals by incorporating a structural causal model (SCM) in an improved variant of Adversarially Learned Inference (ALI), that generates counterfactuals in accordance with the causal relationships between attributes of an image. Based on the generated counterfactuals, we show how to explain a pre-trained machine learning classifier, evaluate its bias, and mitigate the bias using a counterfactual regularizer. On the Morpho-MNIST dataset, our method generates counterfactuals comparable in quality to prior work on SCM-based counterfactuals (DeepSCM), while on the more complex CelebA dataset our method outperforms DeepSCM in generating high-quality valid counterfactuals. Moreover, generated counterfactuals are indistinguishable from reconstructed images in a human evaluation experiment and we subsequently use them to evaluate the fairness of a standard classifier trained on CelebA data. We show that the classifier is biased w.r.t. skin and hair color, and how counterfactual regularization can remove those biases.

Artificial Intelligence, Algorithm Design, and Pricing

Abstract: We calculate the time path of prices generated by algorithmic pricing games that differ in their learning protocols. Asynchronous learning occurs when the algorithm only learns about the return from the action it actually took. Synchronous learning occurs when the artificial intelligence conducts counterfactuals to learn about the returns it would have earned had it taken an alternative action. In a simple market setting, we show that synchronous updating can lead to competitive pricing, while asynchronous updating can lead to pricing close to monopoly levels. However, building simple economic reasoning into the asynchronous algorithms significantly modifies the prices it generates.

Causal learning, counterfactual reasoning and pretend play: a cross-cultural comparison of Peruvian, mixed- and low-socioeconomic status U.S. children

Abstract: Pretend play universally emerges during early childhood and may support the development of causal inference and counterfactual reasoning. However, the amount of time spent pretending, the value that adults place on pretence and the scaffolding adults provide vary by both culture and socioeconomic status (SES). In middle class U.S. preschoolers, accuracy on a pretence-based causal reasoning task predicted performance on a similar causal counterfactual task. We explore the relationship between cultural environment, pretence and counterfactual reasoning in low-income Peruvian (N = 62) and low-income U.S. (N = 57) 3- to 4-year olds, and contrast findings against previous findings in an age-matched, mixed-SES U.S. sample (N = 60). Children learned a novel causal relationship, then answered comparable counterfactual and pretence-based questions about the relationship. Children's responses for counterfactual and pretence measures differed across populations, with Peruvian and lower-income U.S. children providing fewer causally consistent responses when compared with middle class U.S. children. Nevertheless, correlations between the two measures emerged in all populations. Across cohorts, children also provided more causally consistent answers during pretence than counterfactually. Our findings strengthen the hypothesis that causal pretend play is related to causal counterfactual reasoning across cultural contexts, while also suggesting a role for systematic environmental differences. This article is part of the theme issue ‘Thinking about possibilities: mechanisms, ontogeny, functions and phylogeny’.

The Meaning of If

Abstract: Conditional sentences remain a puzzling source of philosophical speculation in large part because there seems to be nothing they could possibly mean that would vindicate the roles they play in language and thought. Bringing together work from philosophy and linguistics, Justin Khoo articulates a theory of what conditionals mean that captures their varied and complex behavior. According to the theory, conditionals form a unified class of expressions that share a common semantic core that encodes inferential dispositions. Thus, rather than represent the world, conditionals are devices used to communicate how we are disposed to infer. Khoo shows that this core theory can be extended to predict the correct probabilities of conditionals, as well as the semantic and pragmatic between different kinds of conditionals. The resulting theory has broad implications beyond debates about the meaning of conditionals, including upshots about the nature of metaphysical and epistemic possibility, the cognitive roles of non-factual contents, and the relationship between counterfactuals and causation.

Counterfactuals of effects of vaccination and public health measures on COVID-19 cases in Canada: What could have happened?

Abstract: This study illustrates what may have happened, in terms of coronavirus disease 2019 (COVID-19) infections, hospitalizations and deaths in Canada, had public health measures not been used to control the COVID-19 epidemic, and had restrictions been lifted with low levels of vaccination, or no vaccination, of the Canadian population. The timeline of the epidemic in Canada, and the public health interventions used to control the epidemic, are reviewed. Comparisons against outcomes in other countries and counterfactual modelling illustrate the relative success of control of the epidemic in Canada. Together, these observations show that without the use of restrictive measures and without high levels of vaccination, Canada could have experienced substantially higher numbers of infections and hospitalizations and almost a million deaths.

Counterfactual Building and Evaluation via eXplainable Support Vector Data Description

Abstract: Increasingly in recent times, the mere prediction of a machine learning algorithm is considered insufficient to gain complete control over the event being predicted. A machine learning algorithm should be considered reliable in the way it allows to extract more knowledge and information than just having a prediction at hand. In this perspective, the counterfactual theory plays a central role. By definition, a counterfactual is the smallest variation of the input such that it changes the predicted behaviour. The paper addresses counterfactuals through Support Vector Data Description (SVDD), empowered by explainability and metric for assessing the counterfactual quality. After showing the specific case in which an analytical solution may be found (under Euclidean distance and linear kernel), an optimisation problem is posed for any type of distances and kernels. The vehicle platooning application is the use case considered to demonstrate how the outlined methodology may offer support to safety-critical applications as well as how explanation may shed new light into the control of the system at hand.

FOCUS: Flexible Optimizable Counterfactual Explanations for Tree Ensembles

Abstract: Model interpretability has become an important problem in machine learning (ML) due to the increased effect algorithmic decisions have on humans. Counterfactual explanations can help users understand not only why ML models make certain decisions, but also how these decisions can be changed. We frame the problem of finding counterfactual explanations as an optimization task and extend previous work that could only be applied to differentiable models. In order to accommodate non-differentiable models such as tree ensembles, we use probabilistic model approximations in the optimization framework. We introduce an approximation technique that is effective for finding counterfactual explanations for predictions of the original model and show that our counterfactual examples are significantly closer to the original instances than those produced by other methods specifically designed for tree ensembles.

Improving Visualization Interpretation Using Counterfactuals

Abstract: Complex, high-dimensional data is used in a wide range of domains to explore problems and make decisions. Analysis of high-dimensional data, however, is vulnerable to the hidden influence of confounding variables, especially as users apply ad hoc filtering operations to visualize only specific subsets of an entire dataset. Thus, visual data-driven analysis can mislead users and encourage mistaken assumptions about causality or the strength of relationships between features. This work introduces a novel visual approach designed to reveal the presence of confounding variables via counterfactual possibilities during visual data analysis. It is implemented in CoFact, an interactive visualization prototype that determines and visualizes counterfactual subsets to better support user exploration of feature relationships. Using publicly available datasets, we conducted a controlled user study to demonstrate the effectiveness of our approach; the results indicate that users exposed to counterfactual visualizations formed more careful judgments about feature-to-outcome relationships.

Explain the Explainer: Interpreting Model-Agnostic Counterfactual Explanations of a Deep Reinforcement Learning Agent

Abstract: Counterfactual examples (CFs) are one of the most popular methods for attaching post-hoc explanations to machine learning (ML) models. However, existing CF generation methods either exploit the internals of specific models or depend on each sample's neighborhood; thus, they are hard to generalize for complex models and inefficient for large datasets. This work aims to overcome these limitations and introduces RELAX, a model-agnostic algorithm to generate optimal counterfactual explanations. Specifically, we formulate the problem of crafting CFs as a sequential decision-making task. We then find the optimal CFs via deep reinforcement learning (DRL) with discretecontinuous hybrid action space. In addition, we develop a distillation algorithm to extract decision rules from the DRL agent's policy in the form of a decision tree to make the process of generating CFs itself interpretable. Extensive experiments conducted on six tabular datasets have shown that RELAX outperforms existing CF generation baselines, as it produces sparser counterfactuals, is more scalable to complex target models to explain, and generalizes to both classification and regression tasks. Finally, we show the ability of our method to provide actionable recommendations and distill interpretable policy explanations in two practical, real-world use cases.