IJCAI 2022 论文列表

Bin packing is a classic optimization problem with a wide range of applications from load balancing to supply chain management. In this work, we study the online variant of the problem, in which a sequence of items of various sizes must be placed into a minimum number of bins of uniform capacity. The online algorithm is enhanced with a (potentially erroneous) prediction concerning the frequency of item sizes in the sequence. We design and analyze online algorithms with efficient tradeoffs between the consistency (i.e., the competitive ratio assuming no prediction error) and the robustness (i.e., the competitive ratio under adversarial error), and whose performance degrades near-optimally as a function of the prediction error. This is the first theoretical and experimental study of online bin packing in the realistic setting of learnable predictions. Previous work addressed only extreme cases with respect to the prediction error, and relied on overly powerful and error-free oracles.

Assume k candidates need to be selected. The candidates appear over time. Each time one appears, it must be immediately selected or rejected---a decision that is made by a group of individuals through voting. Assume the voters use approval ballots, i.e., for each candidate they only specify whether they consider it acceptable or not. This setting can be seen as a voting variant of choosing k secretaries. Our contribution is twofold. (1) We assess to what extent the committees that are computed online can proportionally represent the voters. (2) If a prior probability over candidate approvals is available, we show how to compute committees with maximal expected score.

Semi-weakly supervised semantic segmentation (SWSSS) aims to train a model to identify objects in images based on a small number of images with pixel-level labels, and many more images with only image-level labels. Most existing SWSSS algorithms extract pixel-level pseudo-labels from an image classifier - a very difficult task to do well, hence requiring complicated architectures and extensive hyperparameter tuning on fully-supervised validation sets. We propose a method called prediction filtering, which instead of extracting pseudo-labels, just uses the classifier as a classifier: it ignores any segmentation predictions from classes which the classifier is confident are not present. Adding this simple post-processing method to baselines gives results competitive with or better than prior SWSSS algorithms. Moreover, it is compatible with pseudo-label methods: adding prediction filtering to existing SWSSS algorithms further improves segmentation performance.

Research on human-AI teams usually provides experts with a single label, which ignores the uncertainty in a model's recommendation. Conformal prediction (CP) is a well established line of research that focuses on building a theoretically grounded, calibrated prediction set, which may contain multiple labels. We explore how such prediction sets impact expert decision-making in human-AI teams. Our evaluation on human subjects finds that set valued predictions positively impact experts. However, we notice that the predictive sets provided by CP can be very large, which leads to unhelpful AI assistants. To mitigate this, we introduce D-CP, a method to perform CP on some examples and defer to experts. We prove that D-CP can reduce the prediction set size of non-deferred examples. We show how D-CP performs in quantitative and in human subject experiments (n=120). Our results suggest that CP prediction sets improve human-AI team performance over showing the top-1 prediction alone, and that experts find D-CP prediction sets are more useful than CP prediction sets.

In a coalitional game, the coalitions are weakly ordered according to their worths in the game. When moreover a power index is given, the players are ranked according to the real numbers they are assigned by the power index. If any game inducing the same ordering of the coalitions generates the same ranking of the players then, by definition, the game is (ordinally) stable for the power index, which in turn is ordinally invariant for the game. If one is interested in ranking players of a game which is stable, re-computing the power indices when the coalitional worths slightly fluctuate or are uncertain becomes useless. Bivalued games are easy examples of games stable for any power index which is linear. Among general games, we characterize those that are stable for a given linear index. Note that the Shapley and Banzhaf scores, frequently used in AI, are particular semivalues, and all semivalues are linear indices. To check whether a game is stable for a specific semivalue, it suffices to inspect the ordering of the coalitions and to perform some direct computation based on the semivalue parameters.

Margin has played an important role on the design and analysis of learning algorithms during the past years, mostly working with the maximization of the minimum margin. Recent years have witnessed the increasing empirical studies on the optimization of margin distribution according to different statistics such as medium margin, average margin, margin variance, etc., whereas there is a relative paucity of theoretical understanding. In this work, we take one step on this direction by providing a new generalization error bound, which is heavily relevant to margin distribution by incorporating ingredients such as average margin and semi-variance, a new margin statistics for the characterization of margin distribution. Inspired by the theoretical findings, we propose the MSVMAv, an efficient approach to achieve better performance by optimizing margin distribution in terms of its empirical average margin and semi-variance. We finally conduct extensive experiments to show the superiority of the proposed MSVMAv approach.

Recently, recommender system (RS) based on causal inference has gained much attention in the industrial community, as well as the states of the art performance in many prediction and debiasing tasks. Nevertheless, a unified causal analysis framework has not been established yet. Many causal-based prediction and debiasing studies rarely discuss the causal interpretation of various biases and the rationality of the corresponding causal assumptions. In this paper, we first provide a formal causal analysis framework to survey and unify the existing causal-inspired recommendation methods, which can accommodate different scenarios in RS. Then we propose a new taxonomy and give formal causal definitions of various biases in RS from the perspective of violating the assumptions adopted in causal analysis. Finally, we formalize many debiasing and prediction tasks in RS, and summarize the statistical and machine learning-based causal estimation methods, expecting to provide new research opportunities and perspectives to the causal RS community.

We argue that linear temporal logic LTL in tandem with monadic first-order logic can be used as a ba- sic language for ontology-based access to tempo- ral data and obtain a classification of the resulting ontology-mediated queries according to the type of standard first-order queries they can be rewritten to.

Reward is the driving force for reinforcement-learning agents. We here set out to understand the expressivity of Markov reward as a way to capture tasks that we would want an agent to perform. We frame this study around three new abstract notions of "task": (1) a set of acceptable behaviors, (2) a partial ordering over behaviors, or (3) a partial ordering over trajectories. Our main results prove that while reward can express many of these tasks, there exist instances of each task type that no Markov reward function can capture. We then provide a set of polynomial-time algorithms that construct a Markov reward function that allows an agent to perform each task type, and correctly determine when no such reward function exists.

Fictitious play (FP) is one of the most fundamental game-theoretical learning frameworks for computing Nash equilibrium in n-player games, which builds the foundation for modern multi-agent learning algorithms. Although FP has provable convergence guarantees on zero-sum games and potential games, many real-world problems are often a mixture of both and the convergence property of FP has not been fully studied yet. In this paper, we extend the convergence results of FP to the combinations of such games and beyond. Specifically, we derive new conditions for FP to converge by leveraging game decomposition techniques. We further develop a linear relationship unifying cooperation and competition in the sense that these two classes of games are mutually transferable. Finally, we analyse a non-convergent example of FP, the Shapley game, and develop sufficient conditions for FP to converge.

Model-reconciliation explanation is a popular framework for generating explanations for planning problems. While the framework has been extended to multiple settings since its introduction for classical planning problems, there is little agreement on the computational complexity of generating minimal model reconciliation explanations in the basic setting. In this paper, we address this lacuna by introducing a decision-version of the model-reconciliation explanation generation problem and we show that it is Sigma-2-P Complete.

We consider the problem Enum·IP of enumerating prime implicants of Boolean functions represented by decision decomposable negation normal form (dec-DNNF) circuits. We study Enum·IP from dec-DNNF within the framework of enumeration complexity and prove that it is in OutputP, the class of output polynomial enumeration problems, and more precisely in IncP, the class of polynomial incremental time enumeration problems. We then focus on two closely related, but seemingly harder, enumeration problems where further restrictions are put on the prime implicants to be generated. In the first problem, one is only interested in prime implicants representing subset-minimal abductive explanations, a notion much investigated in AI for more than thirty years. In the second problem, the target is prime implicants representing sufficient reasons, a recent yet important notion in the emerging field of eXplainable AI, since they aim to explain predictions achieved by machine learning classifiers. We provide evidence showing that enumerating specific prime implicants corresponding to subset-minimal abductive explanations or to sufficient reasons is not in OutputP.

We study approval-based multiwinner voting where candidates are in a metric space and committees are valuated in terms of their distances to the given votes. In particular, we consider three different distance functions, and for each of them we study both the utilitarian rules and the egalitarian rules, resulting in six variants of winners determination problems. We focus on the (parameterized) complexity of these problems for both the general metric and several special metrics. For hardness results, we also discuss their approximability.

Network pruning is considered efficient for sparsification and acceleration of Convolutional Neural Network (CNN) based models that can be adopted in re-source-constrained environments. Inspired by two popular pruning criteria, i.e. magnitude and similarity, this paper proposes a novel structural pruning method based on Graph Convolution Network (GCN) to further promote compression performance. The channel features are firstly extracted by Global Average Pooling (GAP) from a batch of samples, and a graph model for each layer is generated based on the similarity of features. A set of agents for individual CNN layers are implemented by GCN and utilized to synthesize comprehensive channel information and determine the pruning scheme for the overall CNN model. The training process of each agent is carried out using Reinforcement Learning (RL) to ensure their convergence and adaptability to various network architectures. The proposed solution is assessed based on a range of image classification datasets i.e., CIFAR and Tiny-ImageNet. The numerical results indicate that the proposed pruning method outperforms the pure magnitude-based or similarity-based pruning solutions and other SOTA methods (e.g., HRank and SCP). For example, the proposed method can prune VGG16 by removing 93% of the model parameters without any accuracy reduction in the CIFAR10 dataset.

In recent years, researchers have made significant progress in devising reinforcement-learning algorithms for optimizing linear temporal logic (LTL) objectives and LTL-like objectives. Despite these advancements, there are fundamental limitations to how well this problem can be solved. Previous studies have alluded to this fact but have not examined it in depth. In this paper, we address the tractability of reinforcement learning for general LTL objectives from a theoretical perspective. We formalize the problem under the probably approximately correct learning in Markov decision processes (PAC-MDP) framework, a standard framework for measuring sample complexity in reinforcement learning. In this formalization, we prove that the optimal policy for any LTL formula is PAC-MDP-learnable if and only if the formula is in the most limited class in the LTL hierarchy, consisting of formulas that are decidable within a finite horizon. Practically, our result implies that it is impossible for a reinforcement-learning algorithm to obtain a PAC-MDP guarantee on the performance of its learned policy after finitely many interactions with an unconstrained environment for LTL objectives that are not decidable within a finite horizon.

Public observation logic (POL) is a variant of dynamic epistemic logic to reason about agent expectations and agent observations. Agents have certain expectations, regarding the situation at hand, that are actuated by the relevant protocols, and they eliminate possible worlds in which their expectations do not match with their observations. In this work, we investigate the computational complexity of the model checking problem for POL and prove its PSPACE-completeness. We also study various syntactic fragments of POL. We exemplify the applicability of POL model checking in verifying different characteristics and features of an interactive system with respect to the distinct expectations and (matching) observations of the system. Finally, we provide a discussion on the implementation of the model checking algorithms.

Dialogue state tracking (DST) is a component of the task oriented dialogue system. It is responsible for extracting and managing slots, where each slot represents a part of the information to accomplish a task, and slot value is updated recurrently in each dialogue turn. However, many DST models cannot update slot values appropriately. These models may repeatedly inherit wrong slot values extracted in previous turns, resulting in the fail of the entire DST task. They cannot update indirectly mentioned slots well, either. This study designed a model with a mentioned slot pool (MSP) to tackle the update problem. The MSP is a slot specific memory that records all mentioned slot values that may be inherited, and our model updates slot values according to the MSP and the dialogue context. Our model rejects inheriting the previous slot value when it predicates the value is wrong. Then, it extracts the slot value from the current dialogue context. As the contextual information accumulates, the new value is more likely to be correct. It also can track the indirectly mentioned slot by picking a value from the MSP. Experimental results showed our model reached state of the art DST performance on MultiWOZ datasets.

Quantified Boolean logic results from adding operators to Boolean logic for existentially and universally quantifying variables. This extends the reach of Boolean logic by enabling a variety of applications that have been explored over the decades. The existential quantification of literals (variable states) and its applications have also been studied in the literature. We complement this by studying universal literal quantification and its applications, particularly to explainable AI. We also provide a novel semantics for quantification and discuss the interplay between variable/literal and existential/universal quantification. We further identify classes of Boolean formulas and circuits that allow efficient quantification. Literal quantification is more fine-grained than variable quantification, which leads to a refinement of quantified Boolean logic with literal quantification as its primitive.

Abductive explanations take a central place in eXplainable Artificial Intelligence (XAI) by clarifying with few features the way data instances are classified. However, instances may have exponentially many minimum-size abductive explanations, and this source of complexity holds even for ``intelligible'' classifiers, such as decision trees. When the number of such abductive explanations is huge, computing one of them, only, is often not informative enough. Especially, better explanations than the one that is derived may exist. As a way to circumvent this issue, we propose to leverage a model of the explainee, making precise her / his preferences about explanations, and to compute only preferred explanations. In this paper, several models are pointed out and discussed. For each model, we present and evaluate an algorithm for computing preferred majoritary reasons, where majoritary reasons are specific abductive explanations suited to random forests. We show that in practice the preferred majoritary reasons for an instance can be far less numerous than its majoritary reasons.