IJCAI 2022 论文列表

Graph-Convolution-based methods have been successfully applied to representation learning on homophily graphs where nodes with the same label or similar attributes tend to connect with one another. Due to the homophily assumption of Graph Convolutional Networks (GCNs) that these methods use, they are not suitable for heterophily graphs where nodes with different labels or dissimilar attributes tend to be adjacent. Several methods have attempted to address this heterophily problem, but they do not change the fundamental aggregation mechanism of GCNs because they rely on summation operators to aggregate information from neighboring nodes, which is implicitly subject to the homophily assumption. Here, we introduce a novel aggregation mechanism and develop a RAndom Walk Aggregation-based Graph Neural Network (called RAW-GNN) method. The proposed approach integrates the random walk strategy with graph neural networks. The new method utilizes breadth-first random walk search to capture homophily information and depth-first search to collect heterophily information. It replaces the conventional neighborhoods with path-based neighborhoods and introduces a new path-based aggregator based on Recurrent Neural Networks. These designs make RAW-GNN suitable for both homophily and heterophily graphs. Extensive experimental results showed that the new method achieved state-of-the-art performance on a variety of homophily and heterophily graphs.

We introduce a Power-of-Two post-training quantization( PTQ) method for deep neural network that meets hardware requirements and does not call for long-time retraining. PTQ requires a small set of calibration data and is easier for deployment, but results in lower accuracy than Quantization-Aware Training( QAT). Power-of-Two quantization can convert the multiplication introduced by quantization and dequantization to bit-shift that is adopted by many efficient accelerators. However, the Power-of-Two scale has fewer candidate values, which leads to more rounding or clipping errors. We propose a novel Power-of-Two PTQ framework, dubbed RAPQ, which dynamically adjusts the Power-of-Two scales of the whole network instead of statically determining them layer by layer. It can theoretically trade off the rounding error and clipping error of the whole network. Meanwhile, the reconstruction method in RAPQ is based on the BN information of every unit. Extensive experiments on ImageNet prove the excellent performance of our proposed method. Without bells and whistles, RAPQ can reach accuracy of 65% and 48% on ResNet-18 and MobileNetV2 respectively with weight INT2 activation INT4. We are the first to propose PTQ for the more constrained but hardware-friendly Power-of-Two quantization and prove that it can achieve nearly the same accuracy as SOTA PTQ method. The code will be released.

Ordinal embedding (OE) aims to project objects into a low-dimensional space while preserving their ordinal constraints as well as possible. Generally speaking, a reasonable OE algorithm should simultaneously capture a) semantic meaning and b) the ordinal relationship of the objects. However, most of the existing methods merely focus on b). To address this issue, our goal in this paper is to seek a generic OE method to embrace the two features simultaneously. We argue that different dimensions of vector-based embedding are naturally entangled with each other. To realize a), we expect to decompose the D dimensional embedding space into D different semantic subspaces, where each subspace is associated with a matrix representation. Unfortunately, introducing a matrix-based representation requires far more complex parametric space than its vector-based counterparts. Thanks to the algebraic property of quaternions, we are able to find a more efficient way to represent a matrix with quaternions. For b), inspired by the classic chordal Grassmannian distance, a new distance function is defined to measure the distance between different quaternions/matrices, on top of which we construct a generic OE loss function. Experimental results for different tasks on both simulated and real-world datasets verify the effectiveness of our proposed method.

AI technologies are being increasingly tested and applied in critical environments including healthcare. Without an effective way to detect and mitigate AI induced inequalities, AI might do more harm than good, potentially leading to the widening of underlying inequalities. This paper proposes a generic allocation-deterioration framework for detecting and quantifying AI induced inequality. Specifically, AI induced inequalities are quantified as the area between two allocation-deterioration curves. To assess the framework’s performance, experiments were conducted on ten synthetic datasets (N>33,000) generated from HiRID - a real-world Intensive Care Unit (ICU) dataset, showing its ability to accurately detect and quantify inequality proportionally to controlled inequalities. Extensive analyses were carried out to quantify health inequalities (a) embedded in two real-world ICU datasets; (b) induced by AI models trained for two resource allocation scenarios. Results showed that compared to men, women had up to 33% poorer deterioration in markers of prognosis when admitted to HiRID ICUs. All four AI models assessed were shown to induce significant inequalities (2.45% to 43.2%) for non-White compared to White patients. The models exacerbated data embedded inequalities significantly in 3 out of 8 assessments, one of which was >9 times worse.

Quantified conflict-driven clause learning (QCDCL) is one of the main approaches for solving quantified Boolean formulas (QBF). We formalise and investigate several versions of QCDCL that include cube learning and/or pure-literal elimination, and formally compare the resulting solving models via proof complexity techniques. Our results show that almost all of the QCDCL models are exponentially incomparable with respect to proof size (and hence solver running time), pointing towards different orthogonal ways how to practically implement QCDCL.

We study signaling in Bayesian ad auctions, in which bidders' valuations depend on a random, unknown state of nature. The auction mechanism has complete knowledge of the actual state of nature, and it can send signals to bidders so as to disclose information about the state and increase revenue. For instance, a state may collectively encode some features of the user that are known to the mechanism only, since the latter has access to data sources unaccessible to the bidders. We study the problem of computing how the mechanism should send signals to bidders in order to maximize revenue. While this problem has already been addressed in the easier setting of second-price auctions, to the best of our knowledge, our work is the first to explore ad auctions with more than one slot. In this paper, we focus on public signaling and VCG mechanisms, under which bidders truthfully report their valuations. We start with a negative result, showing that, in general, the problem does not admit a PTAS unless P = NP, even when bidders' valuations are known to the mechanism. The rest of the paper is devoted to settings in which such negative result can be circumvented. First, we prove that, with known valuations, the problem can indeed be solved in polynomial time when either the number of states d or the number of slots m is fixed. Moreover, in the same setting, we provide an FPTAS for the case in which bidders are single minded, but d and m can be arbitrary. Then, we switch to the random valuations setting, in which these are randomly drawn according to some probability distribution. In this case, we show that the problem admits an FPTAS, a PTAS, and a QPTAS, when, respectively, d is fixed, m is fixed, and bidders' valuations are bounded away from zero.

Depression is a prominent health challenge to the world, and early risk detection (ERD) of depression from online posts can be a promising technique for combating the threat. Early depression detection faces the challenge of efficiently tackling streaming data, balancing the tradeoff between timeliness, accuracy and explainability. To tackle these challenges, we propose a psychiatric scale guided risky post screening method that can capture risky posts related to the dimensions defined in clinical depression scales, and providing interpretable diagnostic basis. A Hierarchical Attentional Network equipped with BERT (HAN-BERT) is proposed to further advance explainable predictions. For ERD, we propose an online algorithm based on an evolving queue of risky posts that can significantly reduce the number of model inferences to boost efficiency. Experiments show that our method outperforms the competitive feature-based and neural models under conventional depression detection settings, and achieves simultaneous improvement in both efficacy and efficiency for ERD.

Knowledge distillation aims to transfer the information by minimizing the cross-entropy between the probabilistic outputs of the teacher and student network. In this work, we propose an alternative distillation objective by maximizing the scoring rule, which quantitatively measures the agreement of a distribution to the reference distribution. We demonstrate that the proper and homogeneous scoring rule exhibits more preferable properties for distillation than the original cross entropy based approach. To that end, we present an efficient implementation of the distillation objective based on a pseudo-spherical scoring rule, which is a family of proper and homogeneous scoring rules. We refer to it as pseudo-spherical knowledge distillation. Through experiments on various model compression tasks, we validate the effectiveness of our method by showing its superiority over the original knowledge distillation. Moreover, together with structural distillation methods such as contrastive representation distillation, we achieve state of the art results in CIFAR100 benchmarks.

Neural architecture search (NAS) and network pruning are widely studied efficient AI techniques, but not yet perfect. NAS performs exhaustive candidate architecture search, incurring tremendous search cost. Though (structured) pruning can simply shrink model dimension, it remains unclear how to decide the per-layer sparsity automatically and optimally. In this work, we revisit the problem of layer-width optimization and propose Pruning-as-Search (PaS), an end-to-end channel pruning method to search out desired sub-network automatically and efficiently. Specifically, we add a depth-wise binary convolution to learn pruning policies directly through gradient descent. By combining the structural reparameterization and PaS, we successfully searched out a new family of VGG-like and lightweight networks, which enable the flexibility of arbitrary width with respect to each layer instead of each stage. Experimental results show that our proposed architecture outperforms prior arts by around 1.0% top-1 accuracy under similar inference speed on ImageNet-1000 classification task. Furthermore, we demonstrate the effectiveness of our width search on complex tasks including instance segmentation and image translation. Code and models are released.

We consider graph representation learning in an unsupervised manner. Graph neural networks use neighborhood aggregation as a core component that results in feature smoothing among nodes in proximity. While successful in various prediction tasks, such a paradigm falls short of capturing nodes' similarities over a long distance, which proves to be important for high-quality learning. To tackle this problem, we strengthen the graph with three types of additional graph views, in which each node is directly linked to a set of nodes with the highest similarity in terms of node features, neighborhood features or local structures. Not restricted by connectivity in the original graph, the generated views provide new and complementary perspectives from which to look at the relationship between nodes. Inspired by the recent success of contrastive learning approaches, we propose a self-supervised method that aims to learn node representations by maximizing the agreement between representations across generated views and the original graph, without the requirement of any label information. We also propose a channel-level contrast approach that greatly reduces computation cost. Extensive experiments on six assortative graphs and three disassortative graphs demonstrate the effectiveness of our approach.

When prototyping AI experiences (AIX), interface designers seek effective ways to support end-user tasks through AI capabilities. However, AI poses challenges to design due to its dynamic behavior in response to training data, end-user data, and feedback. Designers must consider AI's uncertainties and offer adaptations such as explainability, error recovery, and automation vs. human task control. Unfortunately, current prototyping tools assume a black-box view of AI, forcing designers to work with separate tools to explore machine learning models, understand model performance, and align interface choices with model behavior. This introduces friction to rapid and iterative prototyping. We propose Model-Informed Prototyping (MIP), a workflow for AIX design that combines model exploration with UI prototyping tasks. Our system, ProtoAI, allows designers to directly incorporate model outputs into interface designs, evaluate design choices across different inputs, and iteratively revise designs by analyzing model breakdowns.

Gossip protocols are programs that can be used by a group of agents to synchronize what information they have. Namely, assuming each agent holds a secret, the goal of a protocol is to reach a situation in which all agents know all secrets. Distributed epistemic gossip protocols use epistemic formulas in the component programs for the agents. In this paper, we study the simplest classes of such gossip protocols: propositional gossip protocols, in which whether an agent wants to initiate a call depends only on the set of secrets that the agent currently knows. It was recently shown that such a protocol can be correct, i.e., always terminates in a state where all agents know all secrets, only when its communication graph is complete. We show here that this characterization dramatically changes when the usual fairness constraints are imposed on the call scheduler used. Finally, we establish that checking the correctness of a given propositional protocol under a fair scheduler is a coNP-complete problem.

Nested named entity recognition (nested NER) is a fundamental task in natural language processing. Various span-based methods have been proposed to detect nested entities with span representations. However, span-based methods do not consider the relationship between a span and other entities or phrases, which is helpful in the NER task. Besides, span-based methods have trouble predicting long entities due to limited span enumeration length. To mitigate these issues, we present the Propose-and-Refine Network (PnRNet), a two-stage set prediction network for nested NER. In the propose stage, we use a span-based predictor to generate some coarse entity predictions as entity proposals. In the refine stage, proposals interact with each other, and richer contextual information is incorporated into the proposal representations. The refined proposal representations are used to re-predict entity boundaries and classes. In this way, errors in coarse proposals can be eliminated, and the boundary prediction is no longer constrained by the span enumeration length limitation. Additionally, we build multi-scale sentence representations, which better model the hierarchical structure of sentences and provide richer contextual information than token-level representations. Experiments show that PnRNet achieves state-of-the-art performance on four nested NER datasets and one flat NER dataset.

We contribute to the programme of lifting proportionality axioms from the multi-winner voting setting to participatory budgeting. We define novel proportionality axioms for participatory budgeting and test them on known proportionality-driven rules such as Phragmén and Rule X. We investigate logical implications among old and new axioms and provide a systematic overview of proportionality criteria in participatory budgeting.

Data-free knowledge distillation (DFKD) conducts knowledge distillation via eliminating the dependence of original training data, and has recently achieved impressive results in accelerating pre-trained language models. At the heart of DFKD is to reconstruct a synthetic dataset by inverting the parameters of the uncompressed model. Prior DFKD approaches, however, have largely relied on hand-crafted priors of the target data distribution for the reconstruction, which can be inevitably biased and often incompetent to capture the intrinsic distributions. To address this problem, we propose a prompt-based method, termed as PromptDFD, that allows us to take advantage of learned language priors, which effectively harmonizes the synthetic sentences to be semantically and grammatically correct. Specifically, PromptDFD leverages a pre-trained generative model to provide language priors and introduces a reinforced topic prompter to control data synthesis, making the generated samples thematically relevant and semantically plausible, and thus friendly to downstream tasks. As shown in our experiments, the proposed method substantially improves the synthesis quality and achieves considerable improvements on distillation performance. In some cases, PromptDFD even gives rise to results on par with those from the data-driven knowledge distillation with access to the original training data.

In this paper, we propose projected gradient descent (PGD) algorithms for signal estimation from noisy nonlinear measurements. We assume that the unknown signal lies near the range of a Lipschitz continuous generative model with bounded inputs. In particular, we consider two cases when the nonlinear link function is either unknown or known. For unknown nonlinearity, we make the assumption of sub-Gaussian observations and propose a linear least-squares estimator. We show that when there is no representation error, the sensing vectors are Gaussian, and the number of samples is sufficiently large, with high probability, a PGD algorithm converges linearly to a point achieving the optimal statistical rate using arbitrary initialization. For known nonlinearity, we assume monotonicity, and make much weaker assumptions on the sensing vectors and allow for representation error. We propose a nonlinear least-squares estimator that is guaranteed to enjoy an optimal statistical rate. A corresponding PGD algorithm is provided and is shown to also converge linearly to the estimator using arbitrary initialization. In addition, we present experimental results on image datasets to demonstrate the performance of our PGD algorithms.

Multi-agent path finding (MAPF) attracts considerable attention in artificial intelligence community. The task in the standard MAPF is to find discrete paths through which agents can navigate from their starting positions to individual goal positions. The combination of two additional requirements makes the problem computationally challenging: agents must not collide with each other and the paths must be optimal with respect to some objective. Two major approaches to optimal MAPF solving include dedicated search-based methods, and compilation-based methods that reduce a MAPF instance to an instance in a different formalism, for which an efficient solver exists. In this survey, we summarize major compilation-based solvers for MAPF using CSP, SAT, and MILP formalisms. We explain the core ideas of the solvers in a simplified and unified way while preserving the merit making them more accessible for a wider audience.

In this paper, we revisit the problem of Differentially Private Stochastic Convex Optimization (DP-SCO) with heavy-tailed data, where the gradient of the loss function has bounded moments. Instead of the case where the loss function is Lipschitz or each coordinate of the gradient has bounded second moment studied previously, we consider a relaxed scenario where each coordinate of the gradient only has bounded (1+v)-th moment with some v∈(0, 1]. Firstly, we start from the one dimensional private mean estimation for heavy-tailed distributions. We propose a novel robust and private mean estimator which is optimal. Based on its idea, we then extend to the general d-dimensional space and study DP-SCO with general convex and strongly convex loss functions. We also provide lower bounds for these two classes of loss under our setting and show that our upper bounds are optimal up to a factor of O(Poly(d)). To address the high dimensionality issue, we also study DP-SCO with heavy-tailed gradient under some sparsity constraint (DP sparse learning). We propose a new method and show it is also optimal up to a factor of O(s*), where s* is the underlying sparsity of the constraint.

We study a federated learning (FL) framework to effectively train models from scarce and skewly distributed labeled data. We consider a challenging yet practical scenario: a few data sources own a small amount of labeled data, while the rest mass sources own purely unlabeled data. Classical FL requires each client to have enough labeled data for local training, thus is not applicable in this scenario. In this work, we design an effective federated semi-supervised learning framework (FedSSL) to fully leverage both labeled and unlabeled data sources. We establish a unified data space across all participating agents, so that each agent can generate mixed data samples to boost semi-supervised learning (SSL), while keeping data locality. We further show that FedSSL can integrate differential privacy protection techniques to prevent labeled data leakage at the cost of minimum performance degradation. On SSL tasks with as small as 0.17% and 1% of MNIST and CIFAR-10 datasets as labeled data, respectively, our approach can achieve 5-20% performance boost over the state-of-the-art methods.