* Article*:

**Conversational AI ? but where is the I?*** Paper*:

**Task Selection Policies for Multitask Learning*** Paper*:

**Counterfactual Reasoning for Fair Clinical Risk Prediction*** Paper*:

**A Causal Bayesian Networks Viewpoint on Fairness**
**20**
*Saturday*
Jul 2019

Posted Causality

in* Article*:

I remember the first time I saw a computer, it was a Power Macintosh 5260 (with Monkey Island on it). I was around 5 years old and I looked at it as if it belonged to another universe. It did, I was not allowed to get anywhere close to it within a 5 mile radius; it was my older brother’s! That did not stop me. I browsed it for hours. The possibilities of computers were infinite and fuelled by the inspiration of sci-fi worlds the dream of talking machines, machines that can assist humans, think themselves and even have feelings never stopped. I kept dreaming about the possibilities of the future.

* Paper*:

One of the questions that arises when designing models that learn to solve multiple tasks simultaneously is how much of the available training budget should be devoted to each individual task. We refer to any formalized approach to addressing this problem (learned or otherwise) as a task selection policy. In this work we provide an empirical evaluation of the performance of some common task selection policies in a synthetic bandit-style setting, as well as on the GLUE benchmark for natural language understanding. We connect task selection policy learning to existing work on automated curriculum learning and off-policy evaluation, and suggest a method based on counterfactual estimation that leads to improved model performance in our experimental settings.

* Paper*:

The use of machine learning systems to support decision making in healthcare raises questions as to what extent these systems may introduce or exacerbate disparities in care for historically underrepresented and mistreated groups, due to biases implicitly embedded in observational data in electronic health records. To address this problem in the context of clinical risk prediction models, we develop an augmented counterfactual fairness criteria to extend the group fairness criteria of equalized odds to an individual level. We do so by requiring that the same prediction be made for a patient, and a counterfactual patient resulting from changing a sensitive attribute, if the factual and counterfactual outcomes do not differ. We investigate the extent to which the augmented counterfactual fairness criteria may be applied to develop fair models for prolonged inpatient length of stay and mortality with observational electronic health records data. As the fairness criteria is ill-defined without knowledge of the data generating process, we use a variational autoencoder to perform counterfactual inference in the context of an assumed causal graph. While our technique provides a means to trade off maintenance of fairness with reduction in predictive performance in the context of a learned generative model, further work is needed to assess the generality of this approach.

Learning to Rank (LTR) from user interactions is challenging as user feedback often contains high levels of bias and noise. At the moment, two methodologies for dealing with bias prevail in the field of LTR: counterfactual methods that learn from historical data and model user behavior to deal with biases; and online methods that perform interventions to deal with bias but use no explicit user models. For practitioners the decision between either methodology is very important because of its direct impact on end users. Nevertheless, there has never been a direct comparison between these two approaches to unbiased LTR. In this study we provide the first benchmarking of both counterfactual and online LTR methods under different experimental conditions. Our results show that the choice between the methodologies is consequential and depends on the presence of selection bias, and the degree of position bias and interaction noise. In settings with little bias or noise counterfactual methods can obtain the highest ranking performance; however, in other circumstances their optimization can be detrimental to the user experience. Conversely, online methods are very robust to bias and noise but require control over the displayed rankings. Our findings confirm and contradict existing expectations on the impact of model-based and intervention-based methods in LTR, and allow practitioners to make an informed decision between the two methodologies.

* Paper*:

We offer a graphical interpretation of unfairness in a dataset as the presence of an unfair causal path in the causal Bayesian network representing the data-generation mechanism. We use this viewpoint to revisit the recent debate surrounding the COMPAS pretrial risk assessment tool and, more generally, to point out that fairness evaluation on a model requires careful considerations on the patterns of unfairness underlying the training data. We show that causal Bayesian networks provide us with a powerful tool to measure unfairness in a dataset and to design fair models in complex unfairness scenarios.

Marginal structural models (MSM) with inverse probability weighting (IPW) are used to estimate causal effects of time-varying treatments, but can result in erratic finite-sample performance when there is low overlap in covariate distributions across different treatment patterns. Modifications to IPW which target the average treatment effect (ATE) estimand either introduce bias or rely on unverifiable parametric assumptions and extrapolation. This paper extends an alternate estimand, the average treatment effect on the overlap population (ATO) which is estimated on a sub-population with a reasonable probability of receiving alternate treatment patterns in time-varying treatment settings. To estimate the ATO within a MSM framework, this paper extends a stochastic pruning method based on the posterior predictive treatment assignment (PPTA) as well as a weighting analogue to the time-varying treatment setting. Simulations demonstrate the performance of these extensions compared against IPW and stabilized weighting with regard to bias, efficiency and coverage. Finally, an analysis using these methods is performed on Medicare beneficiaries residing across 18,480 zip codes in the U.S. to evaluate the effect of coal-fired power plant emissions exposure on ischemic heart disease hospitalization, accounting for seasonal patterns that lead to change in treatment over time.

**19**
*Friday*
Jul 2019

Posted Causality

in* Paper*:

To understand and explain process behaviour we need to be able to see it, and decide its significance, i.e. be able to tell a story about its behaviours. This paper describes a few of the modelling challenges that underlie monitoring and observation of processes in IT, by human or by software. The topic of the observability of systems has been elevated recently in connection with computer monitoring and tracing of processes for debugging and forensics. It raises the issue of well-known principles of measurement, in bounded contexts, but these issues have been left implicit in the Computer Science literature. This paper aims to remedy this omission, by laying out a simple promise theoretic model, summarizing a long standing trail of work on the observation of distributed systems, based on elementary distinguishability of observations, and classical causality, with history. Three distinct views of a system are sought, across a number of scales, that described how information is transmitted (and lost) as it moves around the system, aggregated into journals and logs.

* Paper*:

Reinforcement learning aims at searching the best policy model for decision making, and has been shown powerful for sequential recommendations. The training of the policy by reinforcement learning, however, is placed in an environment. In many real-world applications, however, the policy training in the real environment can cause an unbearable cost, due to the exploration in the environment. Environment reconstruction from the past data is thus an appealing way to release the power of reinforcement learning in these applications. The reconstruction of the environment is, basically, to extract the casual effect model from the data. However, real-world applications are often too complex to offer fully observable environment information. Therefore, quite possibly there are unobserved confounding variables lying behind the data. The hidden confounder can obstruct an effective reconstruction of the environment. In this paper, by treating the hidden confounder as a hidden policy, we propose a deconfounded multi-agent environment reconstruction (DEMER) approach in order to learn the environment together with the hidden confounder. DEMER adopts a multi-agent generative adversarial imitation learning framework. It proposes to introduce the confounder embedded policy, and use the compatible discriminator for training the policies. We then apply DEMER in an application of driver program recommendation. We firstly use an artificial driver program recommendation environment, abstracted from the real application, to verify and analyze the effectiveness of DEMER. We then test DEMER in the real application of Didi Chuxing. Experiment results show that DEMER can effectively reconstruct the hidden confounder, and thus can build the environment better. DEMER also derives a recommendation policy with a significantly improved performance in the test phase of the real application.

* Paper*:

Saliency map generation techniques are at the forefront of explainable AI literature for a broad range of machine learning applications. Our goal is to question the limits of these approaches on more complex tasks. In this paper we apply Layer-Wise Relevance Propagation (LRP) to a sequence-to-sequence attention model trained on a text summarization dataset. We obtain unexpected saliency maps and discuss the rightfulness of these ‘explanations’. We argue that we need a quantitative way of testing the counterfactual case to judge the truthfulness of the saliency maps. We suggest a protocol to check the validity of the importance attributed to the input and show that the saliency maps obtained sometimes capture the real use of the input features by the network, and sometimes do not. We use this example to discuss how careful we need to be when accepting them as explanation.

* Paper*:

Mutual information is one of the essential building blocks of information theory. Yet, it is only finitely defined for distributions with fast decaying tails on a countable joint alphabet of two random elements. The unboundedness of mutual information over the general class of all distributions on a joint alphabet prevents its potential utility to be fully realized. This is in fact a void in the foundation of information theory that needs to be filled. This article proposes a family of generalized mutual information all of whose members 1) are finitely defined for each and every distribution of two random elements on a joint countable alphabet, except the one by Shannon, and 2) enjoy all utilities of a finite Shannon’s mutual information.

* Paper*:

We provide here a general mathematical framework to model attitudes towards ambiguity which uses the formalism of quantum theory as a ‘purely mathematical formalism, detached from any physical interpretation’. We show that the quantum-theoretic framework enables modelling of the ‘Ellsberg paradox’, but it also successfully applies to more concrete human decision-making (DM) tests involving financial, managerial and medical decisions. In particular, we provide a faithful mathematical representation of various empirical studies which reveal that attitudes of managers towards uncertainty shift from ‘ambiguity seeking’ to ‘ambiguity aversion’, and viceversa, thus exhibiting ‘hope effects’ and ‘fear effects’ in management decisions. The present framework provides a new bold and promising direction towards the development of a unified theory of decisions in the presence of uncertainty.

**15**
*Monday*
Jul 2019

Posted Causality

in* Article*:

Correlation is the first step in finding relationships between quantities and deserves some attention . Correlation is defined as the association between quantities , for eg, the sales might increase when income of people increases.

* Article*:

According to scientific research, creativity and the ability to think on our feet is what helped us survive and evolve into the organisms we are today. The topic is analyzed in great detail by Steven Mithen in his book, ‘Creativity in Human Evolution and Prehistory.’ Creativity and pushing the mind to understand and convey thoughts led to language. Creativity is what kept our species alive and helped us reach new evolutionary heights. With everything our creative minds can produce, why does society pressure us to pursue one thing or one focus of creativity? Why can’t we create just to create or learn new things simply to learn new things? At what point did we decide we need to master every skill we decide to adopt? Why are we looked down upon if we lose interest in the things we’ve found a creative outlet in doing in the past?

* Paper*:

Dynamic Bayesian networks (DBNs) are a general model for stochastic processes with partially observed states. Belief filtering in DBNs is the task of inferring the belief state (i.e. the probability distribution over process states) based on incomplete and uncertain observations. In this article, we explore the idea of accelerating the filtering task by automatically exploiting causality in the process. We consider a specific type of causal relation, called passivity, which pertains to how state variables cause changes in other variables. We present the Passivity-based Selective Belief Filtering (PSBF) method, which maintains a factored belief representation and exploits passivity to perform selective updates over the belief factors. PSBF is evaluated in both synthetic processes and a simulated multi-robot warehouse, where it outperformed alternative filtering methods by exploiting passivity.

* Article*:

After 10 years of ImageNet, AI researchers are digging into the details of test sets and some are asking just how much knowledge has really been created with machine learning.

**14**
*Sunday*
Jul 2019

Posted Causality

in* Article*:

Last month, in a controversial move, San Francisco became the first major city in the United States to ban the sale and distribution of electronic cigarettes. Some experts claim that e-cigs are healthier alternatives to conventional cigarettes and can help smokers quit. Others, though, say the proliferation of e-cigarettes is ensnaring a young generation – who likely wouldn’t have started smoking otherwise – in the grips of nicotine addiction. It’s a classic public health conundrum: How should the medical community and legislators advise patients and protect the public when there’s mixed evidence on the potential harms of a given product? This is the central dilemma facing doctors, politicians, and parents as they struggle to contend with the relatively recent introduction of e-cigs.

* Paper*:

Recent advances in neural networks for content generation enable artificial intelligence (AI) models to generate high-quality media manipulations. Here we report on a randomized experiment designed to study the effect of exposure to media manipulations on over 15,000 individuals’ ability to discern machine-manipulated media. We engineer a neural network to plausibly and automatically remove objects from images, and we deploy this neural network online with a randomized experiment where participants can guess which image out of a pair of images has been manipulated. The system provides participants feedback on the accuracy of each guess. In the experiment, we randomize the order in which images are presented, allowing causal identification of the learning curve surrounding participants’ ability to detect fake content. We find sizable and robust evidence that individuals learn to detect fake content through exposure to manipulated media when provided iterative feedback on their detection attempts. Over a succession of only ten images, participants increase their rating accuracy by over ten percentage points. Our study provides initial evidence that human ability to detect fake, machine-generated content may increase alongside the prevalence of such media online.

* Paper*:

While generic drugs offer a cost-effective alternative to brand name drugs, regulators need a method to assess therapeutic equivalence in a post market setting. We develop such a method in the context of assessing the therapeutic equivalence of immediate release (IM) venlafaxine, based on a large insurance claims dataset provided by OptumLabs\textsuperscript{\textregistered}. To properly address this question, our methodology must deal with issues of non-adherence, secular trends in health outcomes, and lack of treatment overlap due to sharp uptake of the generic once it becomes available. We define, identify (under assumptions) and estimate (using G-computation) a causal effect for a time-to-event outcome by extending regression discontinuity to survival curves. We do not find evidence for a lack of therapeutic equivalence of brand and generic IM venlafaxine.

* Paper*:

Injuries occur when an athlete performs a greater amount of activity (workload) than what their body can absorb. To maximize the positive effects of training while avoiding injuries, athletes and coaches need to determine safe workload levels. The International Olympic Committee has recommended using the acute:chronic workload ratio (ACRatio) to monitor injury risk, and has provided thresholds to minimize risk. However, there are several limitations to the ACRatio which may impact the validity of current recommendations. In this review, we discuss previously published and novel challenges with the ACRatio, and possible strategies to address them. These challenges include 1) formulating the ACRatio as a proportion rather than a measure of change, 2) its use of unweighted averages to measure activity loads, 3) inapplicability of the ACRatio to sports where athletes taper their activity, 4) discretization of the ACRatio prior to model selection, 5) the establishment of the model using sparse data, 6) potential bias in the ACRatio of injured athletes, 7) unmeasured confounding, and 8) application of the ACRatio to subsequent injuries.

* Paper*:

Many real-world human behaviors can be characterized as a sequential decision making processes, such as urban travelers choices of transport modes and routes (Wu et al. 2017). Differing from choices controlled by machines, which in general follows perfect rationality to adopt the policy with the highest reward, studies have revealed that human agents make sub-optimal decisions under bounded rationality (Tao, Rohde, and Corcoran 2014). Such behaviors can be modeled using maximum causal entropy (MCE) principle (Ziebart 2010). In this paper, we define and investigate a general reward trans-formation problem (namely, reward advancement): Recovering the range of additional reward functions that transform the agent’s policy from original policy to a predefined target policy under MCE principle. We show that given an MDP and a target policy, there are infinite many additional reward functions that can achieve the desired policy transformation. Moreover, we propose an algorithm to further extract the additional rewards with minimum ‘cost’ to implement the policy transformation.

* Article*:

Why you need to understand the math in order to understand how pathogens spread.

**14**
*Sunday*
Jul 2019

Posted Causality

in* Paper*:

Overlap between treatment groups is required for nonparametric estimation of causal effects. If a subgroup of subjects always receives (or never receives) a given intervention, we cannot estimate the effect of intervention changes on that subgroup without further assumptions. When overlap does not hold globally, characterizing local regions of overlap can inform the relevance of any causal conclusions for new subjects, and can help guide additional data collection. To have impact, these descriptions must be interpretable for downstream users who are not machine learning experts, such as clinicians. We formalize overlap estimation as a problem of finding minimum volume sets and give a method to solve it by reduction to binary classification with Boolean rules. We also generalize our method to estimate overlap in off-policy policy evaluation. Using data from real-world applications, we demonstrate that these rules have comparable accuracy to black-box estimators while maintaining a simple description. In one case study, we perform a user study with clinicians to evaluate rules learned to describe treatment group overlap in post-surgical opioid prescriptions. In another, we estimate overlap in policy evaluation of antibiotic prescription for urinary tract infections.

* Article*:

We hear this sentence over and over again. But what does that actually mean? This small analysis uncovers this topic with the help of R, and simple regressions, focusing on how alcohol impacts health.

* Library*:

Gives conditional distance correlation and performs the conditional distance correlation sure independence screening procedure for ultrahigh dimensional data. The conditional distance correlation is a novel conditional dependence measurement of two random variables given a third variable. The conditional distance correlation sure independence screening is used for screening variables in ultrahigh dimensional setting.

* Article*:

Artificial intelligence (AI) has undergone a renaissance recently, making major progress in key domains such as vision, language, control, and decision-making. This has been due, in part, to cheap data and cheap compute resources, which have fit the natural strengths of deep learning. However, many defining characteristics of human intelligence, which developed under much different pressures, remain out of reach for current approaches. In particular, generalizing beyond one’s experiences–a hallmark of human intelligence from infancy–remains a formidable challenge for modern AI. The following is part position paper, part review, and part unification. We argue that combinatorial generalization must be a top priority for AI to achieve human-like abilities, and that structured representations and computations are key to realizing this objective. Just as biology uses nature and nurture cooperatively, we reject the false choice between ‘hand-engineering’ and ‘end-to-end’ learning, and instead advocate for an approach which benefits from their complementary strengths. We explore how using relational inductive biases within deep learning architectures can facilitate learning about entities, relations, and rules for composing them. We present a new building block for the AI toolkit with a strong relational inductive bias–the graph network–which generalizes and extends various approaches for neural networks that operate on graphs, and provides a straightforward interface for manipulating structured knowledge and producing structured behaviors. We discuss how graph networks can support relational reasoning and combinatorial generalization, laying the foundation for more sophisticated, interpretable, and flexible patterns of reasoning. As a companion to this paper, we have released an open-source software library for building graph networks, with demonstrations of how to use them in practice.

* Paper*:

Estimating average causal effect (ACE) is useful whenever we want to know the effect of an intervention on a given outcome. In the absence of a randomized experiment, many methods such as stratification and inverse propensity weighting have been proposed to estimate ACE. However, it is hard to know which method is optimal for a given dataset or which hyperparameters to use for a chosen method. To this end, we provide a framework to characterize the loss of a causal inference method against the true ACE, by framing causal inference as a representation learning problem. We show that many popular methods, including back-door methods can be considered as weighting or representation learning algorithms, and provide general error bounds for their causal estimates. In addition, we consider the case when unobserved variables can confound the causal estimate and extend proposed bounds using principles of robust statistics, considering confounding as contamination under the Huber contamination model. These bounds are also estimable; as an example, we provide empirical bounds for the Inverse Propensity Weighting (IPW) estimator and show how the bounds can be used to optimize the threshold of clipping extreme propensity scores. Our work provides a new way to reason about competing estimators, and opens up the potential of deriving new methods by minimizing the proposed error bounds.

We empirically demonstrate that graphical models can be a valuable tool in the identification of mediating variables in causal pathways. We make use of graphical models to elucidate the causal pathway through which the treatment influences the levels of fatigue and weakness in people living with HIV (PLHIV) based on a secondary analysis of a categorical dataset collected in a behavioral clinical trial: is weakness a mediator for the treatment and fatigue, or is fatigue a mediator for the treatment and weakness? Causal mediation analysis could not offer any definite answers to these questions.\\ KEYWORDS: Contingency tables; graphical models; loglinear models; HIV; mediation

**11**
*Thursday*
Jul 2019

Posted Causality

in* Paper*:

The pivotal role that event correlation technology plays in todays applications has lead to the emergence of different families of event correlation approaches with a multitude of specialized correlation semantics, including computation models that support the composition and extension of different semantics. However, type-safe embeddings of extensible and composable event patterns into statically-typed general-purpose programming languages have not been systematically explored so far. Event correlation technology has often adopted well-known and intuitive notations from database queries, for which approaches to type-safe embedding do exist. However, we argue in the paper that these approaches, which are essentially descendants of the work on monadic comprehensions, are not well-suited for event correlations and, thus, cannot without further ado be reused/re-purposed for embedding event patterns. To close this gap we propose PolyJoin, a novel approach to type-safe embedding for fully polyvariadic event patterns with polymorphic correlation semantics. Our approach is based on a tagless final encoding with uncurried higher-order abstract syntax (HOAS) representation of event patterns with n variables, for arbitrary $n \in \mathbb{N}$. Thus, our embedding is defined in terms of the host language without code generation and exploits the host language type system to model and type check the type system of the pattern language. Hence, by construction it impossible to define ill-typed patterns. We show that it is possible to have a purely library-level embedding of event patterns, in the familiar join query notation, which is not restricted to monads. PolyJoin is practical, type-safe and extensible. An implementation of it in pure multicore OCaml is readily usable.

* Paper*:

In statistical data assimilation (SDA) and supervised machine learning (ML), we wish to transfer information from observations to a model of the processes underlying those observations. For SDA, the model consists of a set of differential equations that describe the dynamics of a physical system. For ML, the model is usually constructed using other strategies. In this paper, we develop a systematic formulation based on Monte Carlo sampling to achieve such information transfer. Following the derivation of an appropriate target distribution, we present the formulation based on the standard Metropolis-Hasting (MH) procedure and the Hamiltonian Monte Carlo (HMC) method for performing the high dimensional integrals that appear. To the extensive literature on MH and HMC, we add (1) an annealing method using a hyperparameter that governs the precision of the model to identify and explore the highest probability regions of phase space dominating those integrals, and (2) a strategy for initializing the state space search. The efficacy of the proposed formulation is demonstrated using a nonlinear dynamical model with chaotic solutions widely used in geophysics.

* Library*:

Fits localized GP model for dynamic computer experiments via singular value decomposition of the response matrix Y for large N (the number of observations) using the algorithm proposed by Zhang et al. (2018) . The current version only supports 64-bit architecture.

* Paper*:

We analyze several formalizations of conditional probability and find a new one that encompasses all. Our main result is that a preference relation on random quantities called a plausible preorder induces a coherent conditional expectation; and vice versa, that every coherent function can be extended to a conditional expectation induced by a plausible preorder. The advantages of our approach include a convenient justification of probability laws by the properties of plausible preorders, independence on probability interpretations, or the ability to extend conditional probability to any nonzero condition. In particular, if C is a nonzero condition and \Prob is coherent, then it can be extended so that \Prob(0|C)=0, \Prob(C|C)=1 and \Prob(1|C)=1, no matter whether \Prob(C) is zero or whether it is defined.

* Paper*:

The connectivity of a network conveys information about the dependencies between nodes. We show that this information can be analyzed by measuring the uncertainty (and certainty) contained in paths along nodes and links in a network. Specifically, we derive from first principles a measure known as effective information and describe its behavior in common network models. Networks with higher effective information contain more information within the dependencies between nodes. We show how subgraphs of nodes can be grouped into macro-nodes, reducing the size of a network while increasing its effective information, a phenomenon known as causal emergence. We find that causal emergence is common in simulated and real networks across biological, social, informational, and technological domains. Ultimately, these results show that the emergence of higher scales in networks can be directly assessed, and that these higher scales offer a way to create certainty out of uncertainty.

* Paper*:

Neuroimaging datasets keep growing in size to address increasingly complex medical questions. However, even the largest datasets today alone are too small for training complex machine learning models. A potential solution is to increase sample size by pooling scans from several datasets. In this work, we combine 12,207 MRI scans from 15 studies and show that simple pooling is often ill-advised due to introducing various types of biases in the training data. First, we systematically define these biases. Second, we detect bias by experimentally showing that scans can be correctly assigned to their respective dataset with 73.3% accuracy. Finally, we propose to tell causal from confounding factors by quantifying the extent of confounding and causality in a single dataset using causal inference. We achieve this by finding the simplest graphical model in terms of Kolmogorov complexity. As Kolmogorov complexity is not directly computable, we employ the minimum description length to approximate it. We empirically show that our approach is able to estimate plausible causal relationships from real neuroimaging data.

**10**
*Wednesday*
Jul 2019

Posted Causality

inIn many applications (e.g., anomaly detection and security systems) of smart cities, rare events dominate the importance of the total information of big data collected by Internet of Things (IoTs). That is, it is pretty crucial to explore the valuable information associated with the rare events involved in minority subsets of the voluminous amounts of data. To do so, how to effectively measure the information with importance of the small probability events from the perspective of information theory is a fundamental question. This paper first makes a survey of some theories and models with respect to importance measures and investigates the relationship between subjective or semantic importance and rare events in big data. Moreover, some applications for message processing and data analysis are discussed in the viewpoint of information measures. In addition, based on rare events detection, some open challenges related to information measures, such as smart cities, autonomous driving, and anomaly detection in IoTs, are introduced which can be considered as future research directions.

* Paper*:

We study the independence structure of finitely exchangeable distributions over random vectors and random networks. In particular, we provide necessary and sufficient conditions for an exchangeable vector so that its elements are completely independent or completely dependent. We also provide a sufficient condition for an exchangeable vector so that its elements are marginally independent. We then generalize these results and conditions for exchangeable random networks. In this case, it is demonstrated that the situation is more complex. We show that the independence structure of exchangeable random networks lies in one of six regimes represented by undirected and bidirected independence graphs in graphical model sense. In addition, under certain additional assumptions, we provide necessary and sufficient conditions for the exchangeable network distributions to be faithful to each of these graphs.

* Paper*:

In this work, we propose an introspection technique for deep neural networks that relies on a generative model to instigate salient editing of the input image for model interpretation. Such modification provides the fundamental interventional operation that allows us to obtain answers to counterfactual inquiries, i.e., what meaningful change can be made to the input image in order to alter the prediction. We demonstrate how to reveal interesting properties of the given classifiers by utilizing the proposed introspection approach on both the MNIST and the CelebA dataset.

* Paper*:

Causal estimation relies on separating the variation in the outcome due to the confounders from that due to the treatment. To achieve this separation, practitioners can use external sources of randomness that only influence the treatment called instrumental variables (IVs). Traditional IV-methods rely on structural assumptions that limit the effect that the confounders can have on both outcome and treatment. To relax these assumptions we develop a new estimator called the generalized control-function method (GCFN). GCFN’s first stage called variational decoupling (VDE) recovers the residual variation in the treatment given the IV. In the second stage, GCFN regresses the outcome on the treatment and residual variation to compute the causal effect. We evaluate GCFN on simulated data and on recovering the causal effect of slave export on community trust. We show how VDE can help unify IV-estimators and non-IV-estimators.

* Paper*:

Extreme hydrological events in the Danube river basin may severely impact human populations, aquatic organisms, and economic activity. One often characterizes the joint structure of the extreme events using the theory of multivariate and spatial extremes and its asymptotically justified models. There is interest however in cascading extreme events and whether one event causes another. In this paper, we argue that an improved understanding of the mechanism underlying severe events is achieved by combining extreme value modelling and causal discovery. We construct a causal inference method relying on the notion of the Kolmogorov complexity of extreme conditional quantiles. Tail quantities are derived using multivariate extreme value models and causal-induced asymmetries in the data are explored through the minimum description length principle. Our CausEV, for Causality for Extreme Values, approach uncovers causal relations between summer extreme river discharges in the upper Danube basin and finds significant causal links between the Danube and its Alpine tributary Lech.

* Paper*:

Scientists construct and analyze computational models to understand the world. That understanding comes from efforts to augment, combine, and compare models of related phenomena. We propose SemanticModels.jl, a system that leverages techniques from static and dynamic program analysis to process executable versions of scientific models to perform such metamodeling tasks. By framing these metamodeling tasks as metaprogramming problems, SemanticModels.jl enables writing programs that generate and expand models. To this end, we present a category theory-based framework for defining metamodeling tasks, and extracting semantic information from model implementations, and show how this framework can be used to enhance scientific workflows in a working case study.

**10**
*Wednesday*
Jul 2019

Posted Causality

in* Paper*:

With the increase in the amount of data in many fields, a method to consistently and efficiently decipher relationships within high dimensional data sets is important. Because many modern datasets are high-dimensional, univariate independence tests are not applicable. While many multivariate independence tests have R packages available, the interfaces are inconsistent, most are not available in Python. mgcpy is an extensive Python library that includes many state of the art high-dimensional independence testing procedures using a common interface. The package is easy-to-use and is flexible enough to enable future extensions. This manuscript provides details for each of the tests as well as extensive power and run-time benchmarks on a suite of high-dimensional simulations previously used in different publications. The appendix includes demonstrations of how the user can interact with the package, as well as links and documentation.

* Paper*:

We propose a fast, model agnostic method for finding interpretable counterfactual explanations of classifier predictions by using class prototypes. We show that class prototypes, obtained using either an encoder or through class specific k-d trees, significantly speed up the the search for counterfactual instances and result in more interpretable explanations. We introduce two novel metrics to quantitatively evaluate local interpretability at the instance level. We use these metrics to illustrate the effectiveness of our method on an image and tabular dataset, respectively MNIST and Breast Cancer Wisconsin (Diagnostic). The method also eliminates the computational bottleneck that arises because of numerical gradient evaluation for $\textit{black box}$ models.

* Paper*:

Understanding how ‘black-box’ models arrive at their predictions has sparked significant interest from both within and outside the AI community. Our work focuses on doing this by generating local explanations about individual predictions for tree-based ensembles, specifically Gradient Boosting Decision Trees (GBDTs). Given a correctly predicted instance in the training set, we wish to generate a counterfactual explanation for this instance, that is, the minimal perturbation of this instance such that the prediction flips to the opposite class. Most existing methods for counterfactual explanations are (1) model-agnostic, so they do not take into account the structure of the original model, and/or (2) involve building a surrogate model on top of the original model, which is not guaranteed to represent the original model accurately. There exists a method specifically for random forests; we wish to extend this method for GBDTs. This involves accounting for (1) the sequential dependency between trees and (2) training on the negative gradients instead of the original labels.

* Paper*:

Background: In longitudinal data, it is common to create ‘change scores’ by subtracting measurements taken at baseline from those taken at follow-up, and then to analyse the resulting ‘change’ as the outcome variable. In observational data, this approach can produce misleading causal effect estimates. The present article uses directed acyclic graphs (DAGs) and simple simulations to provide an accessible explanation of why change scores do not estimate causal effects in observational data. Methods: Data were simulated to match three general scenarios where the variable representing measurements of the outcome at baseline was a 1) competing exposure, 2) confounder, or 3) mediator for the total causal effect of the exposure on the variable representing measurements of the outcome at follow-up. Regression coefficients were compared between change-score analyses and DAG-informed analyses. Results: Change-score analyses do not provide meaningful causal effect estimates unless the variable representing measurements of the outcome at baseline is a competing exposure, as in a randomised experiment. Where such variables (i.e. baseline measurements of the outcome) are confounders or mediators, the conclusions drawn from analyses of change scores diverge (potentially substantially) from those of DAG-informed analyses. Conclusions: Future observational studies that seek causal effect estimates should avoid analysing change scores and adopt alternative analytical strategies.

* Paper*:

We introduce Invariant Risk Minimization (IRM), a learning paradigm to estimate invariant correlations across multiple training distributions. To achieve this goal, IRM learns a data representation such that the optimal classifier, on top of that data representation, matches for all training distributions. Through theory and experiments, we show how the invariances learned by IRM relate to the causal structures governing the data and enable out-of-distribution generalization.

* Paper*:

Given two events $A$ and $B$, Bayes’ law is based on the argument that the probability of $A$ given $B$ is proportional to the probability of $B$ given $A$. When probabilities are interpreted in the Bayesian sense, Bayes’ law constitutes a learning algorithm which shows how one can learn from a new observation to improve their belief in a theory that is consistent with that observation. Kolmogorov’s notion of algorithmic information, which is based on the theory of algorithms, proposes an objective measure of the amount of information in a finite string about itself and concludes that for any two finite strings $x$ and $y$, the amount of information in $x$ about $y$ is almost equal to the amount of information in $y$ about $x$. We view this conclusion of Kolmogorov as the algorithmic information version of Bayes’ law. This can be easily demonstrated if one considers the work of Levin on prefix Kolmogorov complexity and then expresses the amount of Kolmogorov mutual information between two finite strings using Solomonoff’s a priori probability.

**09**
*Tuesday*
Jul 2019

Posted Causality

inMany computational models were proposed to extract temporal patterns from clinical time series for each patient and among patient group for predictive healthcare. However, the common relations among patients (e.g., share the same doctor) were rarely considered. In this paper, we represent patients and clinicians relations by bipartite graphs addressing for example from whom a patient get a diagnosis. We then solve for the top eigenvectors of the graph Laplacian, and include the eigenvectors as latent representations of the similarity between patient-clinician pairs into a time-sensitive prediction model. We conducted experiments using real-world data to predict the initiation of first-line treatment for Chronic Lymphocytic Leukemia (CLL) patients. Results show that relational similarity can improve prediction over multiple baselines, for example a 5% incremental over long-short term memory baseline in terms of area under precision-recall curve.

* Paper*:

Confounding bias, missing data, and selection bias are three common obstacles to valid causal inference in the data sciences. Covariate adjustment is the most pervasive technique for recovering casual effects from confounding bias. In this paper, we introduce a covariate adjustment formulation for controlling confounding bias in the presence of missing-not-at-random data and develop a necessary and sufficient condition for recovering causal effects using the adjustment. We also introduce an adjustment formulation for controlling both confounding and selection biases in the presence of missing data and develop a necessary and sufficient condition for valid adjustment. Furthermore, we present an algorithm that lists all valid adjustment sets and an algorithm that finds a valid adjustment set containing the minimum number of variables, which are useful for researchers interested in selecting adjustment sets with desired properties.

* Paper*:

We describe the interface between measure theoretic probability and causal inference by constructing causal models on probability spaces within the potential outcomes framework. We find that measure theory provides a precise and instructive language for causality and that consideration of the probability spaces underlying causal models offers clarity into central concepts of causal inference. By closely studying simple, instructive examples, we demonstrate insights into causal effects, causal interactions, matching procedures, and randomization. Additionally, we introduce a simple technique for visualizing causal models on probability spaces that is useful both for generating examples and developing causal intuition. Finally, we provide an axiomatic framework for causality and make initial steps towards a formal theory of general causal models.

* Paper*:

The focus of this paper is on intrinsic methods to detect overfitting. These rely only on the model and the training data, as opposed to traditional extrinsic methods that rely on performance on a test set or on bounds from model complexity. We propose a family of intrinsic methods called Counterfactual Simulation (CFS) which analyze the flow of training examples through the model by identifying and perturbing rare patterns. By applying CFS to logic circuits we get a method that has no hyper-parameters and works uniformly across different types of models such as neural networks, random forests and lookup tables. Experimentally, CFS can separate models with different levels of overfit using only their logic circuit representations without any access to the high level structure. By comparing lookup tables, neural networks, and random forests using CFS, we get insight into why neural networks generalize. In particular, we find that stochastic gradient descent in neural nets does not lead to ‘brute force’ memorization, but finds common patterns (whether we train with actual or randomized labels), and neural networks are not unlike forests in this regard. Finally, we identify a limitation with our proposal that makes it unsuitable in an adversarial setting, but points the way to future work on robust intrinsic methods.

* Paper*:

We extend two kinds of causal models, structural equation models and simulation models, to infinite variable spaces. This enables a semantics for conditionals founded on a calculus of intervention, and axiomatization of causal reasoning for rich, expressive generative models—including those in which a causal representation exists only implicitly—in an open-universe setting. Further, we show that under suitable restrictions the two kinds of models are equivalent, perhaps surprisingly as their axiomatizations differ substantially in the general case. We give a series of complete axiomatizations in which the open-universe nature of the setting is seen to be essential.

* Paper*:

Covariate adjustment is commonly used for total causal effect estimation. In recent years, graphical criteria have been developed to identify all covariate sets that can be used for this purpose. Different valid adjustment sets typically provide causal effect estimates of varying accuracies. We introduce a graphical criterion to compare the asymptotic variance provided by certain valid adjustment sets in a causal linear model. We employ this result to develop two further graphical tools. First, we introduce a simple variance reducing pruning procedure for any given valid adjustment set. Second, we give a graphical characterization of a valid adjustment set that provides the optimal asymptotic variance among all valid adjustment sets. Our results depend only on the graphical structure and not on the specific error variances or the edge coefficients of the underlying causal linear model. They can be applied to DAGs, CPDAGs and maximally oriented PDAGs. We present simulations and a real data example to support our results and show their practical applicability.

**07**
*Sunday*
Jul 2019

Posted Causality

in* Paper*:

There are many AI tasks involving multiple interacting agents where agents should learn to cooperate and collaborate to effectively perform the task. Here we develop and evaluate various multi-agent protocols to train agents to collaborate with teammates in grid soccer. We train and evaluate our multi-agent methods against a team operating with a smart hand-coded policy. As a baseline, we train agents concurrently and independently, with no communication. Our collaborative protocols were parameter sharing, coordinated learning with communication, and counterfactual policy gradients. Against the hand-coded team, the team trained with parameter sharing and the team trained with coordinated learning performed the best, scoring on 89.5% and 94.5% of episodes respectively when playing against the hand-coded team. Against the parameter sharing team, with adversarial training the coordinated learning team scored on 75% of the episodes, indicating it is the most adaptable of our methods. The insights gained from our work can be applied to other domains where multi-agent collaboration could be beneficial.

* Paper*:

Suppose X and Y are binary exposure and outcome variables, and we have full knowledge of the distribution of Y, given application of X. From this we know the average causal effect of X on Y. We are now interested in assessing, for a case that was exposed and exhibited a positive outcome, whether it was the exposure that caused the outcome. The relevant ‘probability of causation’, PC, typically is not identified by the distribution of Y given X, but bounds can be placed on it, and these bounds can be improved if we have further information about the causal process. Here we consider cases where we know the probabilistic structure for a sequence of complete mediators between X and Y. We derive a general formula for calculating bounds on PC for any pattern of data on the mediators (including the case with no data). We show that the largest and smallest upper and lower bounds that can result from any complete mediation process can be obtained in processes with at most two steps. We also consider homogeneous processes with many mediators. PC can sometimes be identified as 0 with negative data, but it cannot be identified at 1 even with positive data on an infinite set of mediators. The results have implications for learning about causation from knowledge of general processes and of data on cases.

* Paper*:

This manuscript introduces an \proglang{R} package called \pkg{trialr} that implements a collection of clinical trial methods in \proglang{Stan} and \proglang{R}. In this article, we explore three methods in detail. The first is the continual reassessment method for conducting phase I dose-finding trials that seek a maximum tolerable dose. The second is EffTox, a dose-finding design that scrutinises doses by joint efficacy and toxicity outcomes. The third is the augmented binary method for modelling the probability of treatment success in phase II oncology trials with reference to repeated measures of continuous tumour size and binary indicators of treatment failure. We emphasise in this article the benefits that stem from having access to posterior samples, including flexible inference and powerful visualisation. We hope that this package encourages the use of Bayesian methods in clinical trials.

* Paper*:

Training convolutional neural networks for image classification tasks usually causes information loss. Although most of the time the information lost is redundant with respect to the target task, there are still cases where discriminative information is also discarded. For example, if the samples that belong to the same category have multiple correlated features, the model may only learn a subset of the features and ignore the rest. This may not be a problem unless the classification in the test set highly depends on the ignored features. We argue that the discard of the correlated discriminative information is partially caused by the fact that the minimization of the classification loss doesn’t ensure to learn the overall discriminative information but only the most discriminative information. To address this problem, we propose an information flow maximization (IFM) loss as a regularization term to find the discriminative correlated features. With less information loss the classifier can make predictions based on more informative features. We validate our method on the shiftedMNIST dataset and show the effectiveness of IFM loss in learning representative and discriminative features.

* Paper*:

Causal approaches to fairness have seen substantial recent interest, both from the machine learning community and from wider parties interested in ethical prediction algorithms. In no small part, this has been due to the fact that causal models allow one to simultaneously leverage data and expert knowledge to remove discriminatory effects from predictions. However, one of the primary assumptions in causal modeling is that you know the causal graph. This introduces a new opportunity for bias, caused by misspecifying the causal model. One common way for misspecification to occur is via unmeasured confounding: the true causal effect between variables is partially described by unobserved quantities. In this work we design tools to assess the sensitivity of fairness measures to this confounding for the popular class of non-linear additive noise models (ANMs). Specifically, we give a procedure for computing the maximum difference between two counterfactually fair predictors, where one has become biased due to confounding. For the case of bivariate confounding our technique can be swiftly computed via a sequence of closed-form updates. For multivariate confounding we give an algorithm that can be efficiently solved via automatic differentiation. We demonstrate our new sensitivity analysis tools in real-world fairness scenarios to assess the bias arising from confounding.

* Paper*:

Complex system arises as a result of the nonlinear interactions between components. In particular, the evolutionary dynamics of a multivariate system encodes the ways in which different variables interact with each other individually or in groups. One fundamental question that remains unanswered is: how do two non-overlapping multivariate subsets of variables interact to causally determine the outcome of a specific variable? Here we provide an information based approach to address this problem. We delineate the temporal interactions between the bundles in a probabilistic graphical model. The strength of the interactions, captured by partial information decomposition, then exposes complex behavior of dependencies and memory within the system. The proposed approach successfully illustrates complex dependence between cations and anions as determinants of \textit{pH} in an observed stream chemistry system. This example demonstrates the potentially broad applicability of the approach, establishing the foundation to study the interaction between groups of variables in a range of complex systems.