| Total: 15
We introduce a simple yet expressive image generation method. On the one hand, it does not require the user to paint the masks or define a bounding box of the various objects, since the model does it by itself. On the other hand, it supports defining a coarse location and size of each object. Based on this, we offer a simple, interactive GUI, that allows a layman user to generate diverse images effortlessly. From a technical perspective, we introduce a dual embedding of layout and appearance. In this scheme, the location, size, and appearance of an object can change independently of each other. This way, the model is able to generate innumerable images per scene graph, to better express the intention of the user. In comparison to previous work, we also offer better quality and higher resolution outputs. This is due to a superior architecture, which is based on a novel set of discriminators. Those discriminators better constrain the shape of the generated mask, as well as capturing the appearance encoding in a counterfactual way. Our code is publicly available at https://www.github.com/ashual/scene_generation.
A key feature of inductive logic programming (ILP) is its ability to learn first-order programs, which are intrinsically more expressive than propositional programs. In this paper, we introduce ILP techniques to learn higher-order programs. We implement our idea in Metagolho, an ILP system which can learn higher-order programs with higher-order predicate invention. Our experiments show that, compared to first-order programs, learning higher-order programs can significantly improve predictive accuracies and reduce learning times.
In this work we have investigated the concept of “restraining bolt”, inspired by Science Fiction. We have two distinct sets of features extracted from the world, one by the agent and one by the authority imposing some restraining specifications on the behaviour of the agent (the “restraining bolt”). The two sets of features and, hence the model of the world attainable from them, are apparently unrelated since of interest to independent parties. However they both account for (aspects of) the same world. We have considered the case in which the agent is a reinforcement learning agent on a set of low-level (subsymbolic) features, while the restraining bolt is specified logically using linear time logic on finite traces f/f over a set of high-level symbolic features. We show formally, and illustrate with examples, that, under general circumstances, the agent can learn while shaping its goals to suitably conform (as much as possible) to the restraining bolt specifications.1
We introduce a new framework for conceiving of and studying algorithms that are deployed to aid human decision making: “algorithm-in-the-loop” systems. The algorithm-in-the-loop framework centers human decision making, providing a more precise lens for studying the social impacts of algorithmic decision making aids. We report on two experiments that evaluate algorithm-in-the-loop decision making and find significant limits to these systems.
We present an approach to explain the decisions of black box image classifiers through synthetic exemplar and counter-exemplar learnt in the latent feature space. Our explanation method exploits the latent representations learned through an adversarial autoencoder for generating a synthetic neighborhood of the image for which an explanation is required. A decision tree is trained on a set of images represented in the latent space, and its decision rules are used to generate exemplar images showing how the original image can be modified to stay within its class. Counterfactual rules are used to generate counter-exemplars showing how the original image can “morph” into another class. The explanation also comprehends a saliency map highlighting the areas that contribute to its classification, and areas that push it into another class. A wide and deep experimental evaluation proves that the proposed method outperforms existing explainers in terms of fidelity, relevance, coherence, and stability, besides providing the most useful and interpretable explanations.
Content moderation, the AI-human hybrid process of removing (toxic) content from social media to promote community health, has attracted increasing attention from lawmakers due to allegations of political bias. Hitherto, this allegation has been made based on anecdotes rather than logical reasoning and empirical evidence, which motivates us to audit its validity. In this paper, we first introduce two formal criteria to measure bias (i.e., independence and separation) and their contextual meanings in content moderation, and then use YouTube as a lens to investigate if the political leaning of a video plays a role in the moderation decision for its associated comments. Our results show that when justifiable target variables (e.g., hate speech and extremeness) are controlled with propensity scoring, the likelihood of comment moderation is equal across left- and right-leaning videos.
Machine learning is often used to produce decision-making rules that classify or evaluate individuals. When these individuals have incentives to be classified a certain way, they may behave strategically to influence their outcomes. We develop a model for how strategic agents can invest effort to change the outcomes they receive, and we give a tight characterization of when such agents can be incentivized to invest specified forms of effort into improving their outcomes as opposed to “gaming” the classifier. We show that whenever any “reasonable” mechanism can do so, a simple linear mechanism suffices. This work is based on “How Do Classifiers Induce Agents To Invest Effort Strategically?” published in Economics and Computation 2019 (Kleinberg and Raghavan 2019).
We study the problem of causal identification from an arbitrary collection of observational and experimental distributions, and substantive knowledge about the phenomenon under investigation, which usually comes in the form of a causal graph. We call this problem g-identifiability, or gID for short. In this paper, we introduce a general strategy to prove non-gID based on thickets and hedgelets, which leads to a necessary and sufficient graphical condition for the corresponding decision problem. We further develop a procedure for systematically computing the target effect, and prove that it is sound and complete for gID instances. In other words, the failure of the algorithm in returning an expression implies that the target effect is not computable from the available distributions. Finally, as a corollary of these results, we show that do-calculus is complete for the task of g-identifiability.
The goal of the unsupervised learning of disentangled representations is to separate the independent explanatory factors of variation in the data without access to supervision. In this paper, we summarize the results of (Locatello et al. 2019b) and focus on their implications for practitioners. We discuss the theoretical result showing that the unsupervised learning of disentangled representations is fundamentally impossible without inductive biases and the practical challenges it entails. Finally, we comment on our experimental findings, highlighting the limitations of state-of-the-art approaches and directions for future research.
Constraint Programming (CP) is a powerful paradigm for solving combinatorial problems. In CP, the user creates a model by declaring variables with their domains and expresses the constraints that need to be satisfied in any solution. The solver is then in charge of finding feasible solutions—a value in the domain of each variable that satisfies all the constraints. The discovery of solutions is done by exploring a search tree that is pruned by the constraints in charge of removing impossible values. The CP framework has the advantage of exposing a rich high-level declarative constraint language for modeling, as well as efficient purpose-specific filtering algorithms that can be reused in many problems. In this work, we harness this flexibility and efficiency for the Block Modeling problem. It is a variant of the graph clustering problem that has been used extensively in many domains including social science, spatio-temporal data analysis and even medical imaging. We present a new approach based on constraint programming, allowing discrete optimization of block modeling in a manner that is not only scalable, but also allows the easy incorporation of constraints. We introduce a new constraint filtering algorithm that outperforms earlier approaches. We show its use in the analysis of real datasets.
The St. Petersburg paradox is a centuries-old puzzle concerning a lottery with infinite expected payoff on which people are only willing to pay a small amount to play. Despite many attempts and several proposals, no generally-accepted resolution is yet at hand. In a recent paper, we show that this paradox can be understood in terms of the mind optimally using its limited computational resources (Nobandegani et al. 2019). Specifically, we show that the St. Petersburg paradox can be accounted for by a variant of normative expected-utility valuation which acknowledges cognitive limitations: sample-based expected utility (Nobandegani et al. 2018). SbEU provides a unified, algorithmic explanation of major experimental findings on this paradox. We conclude by discussing the implications of our work for algorithmically understanding human cognition and for developing human-like artificial intelligence.
The field of artificial intelligence has experienced a dramatic methodological shift towards large neural networks trained on plentiful data. This shift has been fueled by recent advances in hardware and techniques enabling remarkable levels of computation, resulting in impressive advances in AI across many applications. However, the massive computation required to obtain these exciting results is costly both financially, due to the price of specialized hardware and electricity or cloud compute time, and to the environment, as a result of non-renewable energy used to fuel modern tensor processing hardware. In a paper published this year at ACL, we brought this issue to the attention of NLP researchers by quantifying the approximate financial and environmental costs of training and tuning neural network models for NLP (Strubell, Ganesh, and McCallum 2019). In this extended abstract, we briefly summarize our findings in NLP, incorporating updated estimates and broader information from recent related publications, and provide actionable recommendations to reduce costs and improve equity in the machine learning and artificial intelligence community.
This abstract looks at one version of the pathfinding problem in games and discusses how it motived our recent work at the AIIDE 2019 conference.
All known SAT-solving paradigms (backtracking, local search, and the polynomial method) only yield a 2n(1−1/O(k)) time algorithm for solving k-SAT in the worst case, where the big-O constant is independent of k. For this reason, it has been hypothesized that k-SAT cannot be solved in worst-case 2n(1−f(k)/k) time, for any unbounded ƒ : ℕ → ℕ. This hypothesis has been called the “Super-Strong Exponential Time Hypothesis” (Super Strong ETH), modeled after the ETH and the Strong ETH. We prove two results concerning the Super-Strong ETH: 1. It has also been hypothesized that k-SAT is hard to solve for randomly chosen instances near the “critical threshold”, where the clause-to-variable ratio is 2k ln 2 −Θ(1). We give a randomized algorithm which refutes the Super-Strong ETH for the case of random k-SAT and planted k-SAT for any clause-to-variable ratio. In particular, given any random k-SAT instance F with n variables and m clauses, our algorithm decides satisfiability for F in 2n(1−Ω( log k)/k) time, with high probability (over the choice of the formula and the randomness of the algorithm). It turns out that a well-known algorithm from the literature on SAT algorithms does the job: the PPZ algorithm of Paturi, Pudlak, and Zane (1998). 2. The Unique k-SAT problem is the special case where there is at most one satisfying assignment. It is natural to hypothesize that the worst-case (exponential-time) complexity of Unique k-SAT is substantially less than that of k-SAT. Improving prior reductions, we show the time complexities of Unique k-SAT and k-SAT are very tightly related: if Unique k-SAT is in 2n(1−f(k)/k) time for an unbounded f, then k-SAT is in 2n(1−f(k)(1−ɛ)/k) time for every ɛ > 0. Thus, refuting Super Strong ETH in the unique solution case would refute Super Strong ETH in general.
Cardinal scores collected from people are well known to suffer from miscalibrations. A popular approach to address this issue is to assume simplistic models of miscalibration (such as linear biases) to de-bias the scores. This approach, however, often fares poorly because people's miscalibrations are typically far more complex and not well understood. It is widely believed that in the absence of simplifying assumptions on the miscalibration, the only useful information in practice from the cardinal scores is the induced ranking. In this paper we address the fundamental question of whether this widespread folklore belief is actually true. We consider cardinal scores with arbitrary (or even adversarially chosen) miscalibrations that is only required to be consistent with the induced ranking. We design rating-based estimators and prove that despite making no assumptions on the ratings, they strictly and uniformly outperform all possible estimators that rely on only the ranking. These estimators can be used as a plug-in to show the superiority of cardinal scores over ordinal rankings for a variety of applications, including A/B testing and ranking. This work thus provides novel fundamental insights in the eternal debate between cardinal and ordinal data: It ranks the approach of using ratings higher than that of using rankings, and rates both approaches in terms of their estimation errors.