AAAI 2023 论文列表

In this paper, we measure the privacy leakage via studying whether graph representations can be inverted to recover the graph used to generate them via graph reconstruction attack (GRA). We propose a GRA that recovers a graph's adjacency matrix from the representations via a graph decoder that minimizes the reconstruction loss between the partial graph and the reconstructed graph. We study three types of representations that are trained on the graph, i.e., representations output from graph convolutional network (GCN), graph attention network (GAT), and our proposed simplicial neural network (SNN) via a higher-order combinatorial Laplacian. Unlike the first two types of representations that only encode pairwise relationships, the third type of representation, i.e., SNN outputs, encodes higher-order interactions (e.g., homological features) between nodes. We find that the SNN outputs reveal the lowest privacy-preserving ability to defend the GRA, followed by those of GATs and GCNs, which indicates the importance of building more private representations with higher-order node information that could defend the potential threats, such as GRAs.

Deep anomaly detection methods learn representations that separate between normal and anomalous images. Although self-supervised representation learning is commonly used, small dataset sizes limit its effectiveness. It was previously shown that utilizing external, generic datasets (e.g. ImageNet classification) can significantly improve anomaly detection performance. One approach is outlier exposure, which fails when the external datasets do not resemble the anomalies. We take the approach of transferring representations pre-trained on external datasets for anomaly detection. Anomaly detection performance can be significantly improved by fine-tuning the pre-trained representations on the normal training images. In this paper, we first demonstrate and analyze that contrastive learning, the most popular self-supervised learning paradigm cannot be naively applied to pre-trained features. The reason is that pre-trained feature initialization causes poor conditioning for standard contrastive objectives, resulting in bad optimization dynamics. Based on our analysis, we provide a modified contrastive objective, the Mean-Shifted Contrastive Loss. Our method is highly effective and achieves a new state-of-the-art anomaly detection performance including 98.6% ROC-AUC on the CIFAR-10 dataset.

Even though data is abundant, it is often subjected to some form of censoring or truncation which inherently creates biases. Removing such biases and performing parameter estimation is a classical challenge in Statistics. In this paper, we focus on the problem of estimating the means of a mixture of two balanced d-dimensional Gaussians when the samples are prone to truncation. A recent theoretical study on the performance of the Expectation-Maximization (EM) algorithm for the aforementioned problem showed EM almost surely converges for d=1 and exhibits local convergence for d>1 to the true means. Nevertheless, the EM algorithm for the case of truncated mixture of two Gaussians is not easy to implement as it requires solving a set of nonlinear equations at every iteration which makes the algorithm impractical. In this work, we propose a gradient based variant of the EM algorithm that has global convergence guarantees when d=1 and local convergence for d>1 to the true means. Moreover, the update rule at every iteration is easy to compute which makes the proposed method practical. We also provide numerous experiments to obtain more insights into the effect of truncation on the convergence to the true parameters in high dimensions.

We study the problem of training a Reinforcement Learning (RL) agent that is collaborative with humans without using human data. Although such agents can be obtained through self-play training, they can suffer significantly from the distributional shift when paired with unencountered partners, such as humans. In this paper, we propose Maximum Entropy Population-based training (MEP) to mitigate such distributional shift. In MEP, agents in the population are trained with our derived Population Entropy bonus to promote the pairwise diversity between agents and the individual diversity of agents themselves. After obtaining this diversified population, a common best agent is trained by paring with agents in this population via prioritized sampling, where the prioritization is dynamically adjusted based on the training progress. We demonstrate the effectiveness of our method MEP, with comparison to Self-Play PPO (SP), Population-Based Training (PBT), Trajectory Diversity (TrajeDi), and Fictitious Co-Play (FCP) in both matrix game and Overcooked game environments, with partners being human proxy models and real humans. A supplementary video showing experimental results is available at https://youtu.be/Xh-FKD0AAKE.

The Voter model is a well-studied stochastic process that models the invasion of a novel trait A (e.g., a new opinion, social meme, genetic mutation, magnetic spin) in a network of individuals (agents, people, genes, particles) carrying an existing resident trait B. Individuals change traits by occasionally sampling the trait of a neighbor, while an invasion bias δ ≥ 0 expresses the stochastic preference to adopt the novel trait A over the resident trait B. The strength of an invasion is measured by the probability that eventually the whole population adopts trait A, i.e., the fixation probability. In more realistic settings, however, the invasion bias is not ubiquitous, but rather manifested only in parts of the network. For instance, when modeling the spread of a social trait, the invasion bias represents localized incentives. In this paper, we generalize the standard biased Voter model to the positional Voter model, in which the invasion bias is effectuated only on an arbitrary subset of the network nodes, called biased nodes. We study the ensuing optimization problem, which is, given a budget k, to choose k biased nodes so as to maximize the fixation probability of a randomly occurring invasion. We show that the problem is NP-hard both for finite δ and when δ → ∞ (strong bias), while the objective function is not submodular in either setting, indicating strong computational hardness. On the other hand, we show that, when δ → 0 (weak bias), we can obtain a tight approximation in O(n^2ω ) time, where ω is the matrix-multiplication exponent. We complement our theoretical results with an experimental evaluation of some proposed heuristics.

Modern social networks are dynamic in their nature; a new connections are appearing and old connections are disappearing all the time. However, in our algorithmic and complexity studies, we usually model social networks as static graphs. In this paper, we propose a new paradigm for the study of the well-known Target Set Selection problem, which is a fundamental problem in viral marketing and the spread of opinion through social networks. In particular, we use temporal graphs to capture the dynamic nature of social networks. We show that the temporal interpretation is, unsurprisingly, NP-complete in general. Then, we study computational complexity of this problem for multiple restrictions of both the threshold function and the underlying graph structure and provide multiple hardness lower-bounds.

DatalogMTL is a powerful extension of Datalog with operators from metric temporal logic (MTL), which has received significant attention in recent years. In this paper, we investigate materialisation-based reasoning (a.k.a. forward chaining) in the context of DatalogMTL programs and datasets with bounded intervals, where partial representations of the canonical model are obtained through successive rounds of rule applications. Although materialisation does not naturally terminate in this setting, it is known that the structure of canonical models is ultimately periodic. Our first contribution in this paper is a detailed analysis of the periodic structure of canonical models; in particular, we formulate saturation conditions whose satisfaction by a partial materialisation implies an ability to recover the full canonical model via unfolding; this allows us to compute the actual periods describing the repeating parts of the canonical model as well as to establish concrete bounds on the number of rounds of rule applications required to achieve saturation. Based on these theoretical results, we propose a practical reasoning algorithm where saturation can be efficiently detected as materialisation progresses, and where the relevant periods used to evaluate entailment of queries via unfolding are efficiently computed. We have implemented our algorithm and our experiments suggest that our approach is both scalable and robust.

The present paper introduces sparsely supervised instance segmentation, with the datasets being fully annotated bounding boxes and sparsely annotated masks. A direct solution to this task is self-training, which is not fully explored for instance segmentation yet. In this paper, we propose MaskBooster for sparsely supervised instance segmentation (SpSIS) with comprehensive usage of pseudo masks. MaskBooster is featured with (1) dynamic and progressive pseudo masks from an online updating teacher model, (2) refining binary pseudo masks with the help of bounding box prior, (3) learning inter-class prediction distribution via knowledge distillation for soft pseudo masks. As an end-to-end and universal self-training framework, MaskBooster can empower fully supervised algorithms and boost their segmentation performance on SpSIS. Abundant experiments are conducted on COCO and BDD100K datasets and validate the effectiveness of MaskBooster. Specifically, on different COCO protocols and BDD100K, we surpass sparsely supervised baseline by a large margin for both Mask RCNN and ShapeProp. MaskBooster on SpSIS also outperforms weakly and semi-supervised instance segmentation state-of-the-art on the datasets with similar annotation budgets.

Training a robust system, e.g., Speech to Text (STT), requires large datasets. Variability present in the dataset, such as unwanted nuances and biases, is the reason for the need for large datasets to learn general representations. In this work, we propose a novel approach to induce invariance using adversarial forgetting (AF). Our initial experiments on learning invariant features such as accent on the STT task achieve better generalizations in terms of word error rate (WER) compared to traditional models. We observe an absolute improvement of 2.2% and 1.3% on out-of-distribution and in-distribution test sets, respectively.

Canonical models of Markov decision processes (MDPs) usually consider geometric discounting based on a constant discount factor. While this standard modeling approach has led to many elegant results, some recent studies indicate the necessity of modeling time-varying discounting in certain applications. This paper studies a model of infinite-horizon MDPs with time-varying discount factors. We take a game-theoretic perspective – whereby each time step is treated as an independent decision maker with their own (fixed) discount factor – and we study the subgame perfect equilibrium (SPE) of the resulting game as well as the related algorithmic problems. We present a constructive proof of the existence of an SPE and demonstrate the EXPTIME-hardness of computing an SPE. We also turn to the approximate notion of epsilon-SPE and show that an epsilon-SPE exists under milder assumptions. An algorithm is presented to compute an epsilon-SPE, of which an upper bound of the time complexity, as a function of the convergence property of the time-varying discount factor, is provided.

Although the prevention of AI vulnerabilities is critical to preserve the safety and privacy of users and businesses, educational tools for robust AI are still underdeveloped worldwide. We present the design, implementation, and assessment of Maestro. Maestro is an effective open-source game-based platform that contributes to the advancement of robust AI education. Maestro provides "goal-based scenarios" where college students are exposed to challenging life-inspired assignments in a "competitive programming" environment. We assessed Maestro's influence on students' engagement, motivation, and learning success in robust AI. This work also provides insights into the design features of online learning tools that promote active learning opportunities in the robust AI domain. We analyzed the reflection responses (measured with Likert scales) of 147 undergraduate students using Maestro in two quarterly college courses in AI. According to the results, students who felt the acquisition of new skills in robust AI tended to appreciate highly Maestro and scored highly on material consolidation, curiosity, and maestry in robust AI. Moreover, the leaderboard, our key gamification element in Maestro, has effectively contributed to students' engagement and learning. Results also indicate that Maestro can be effectively adapted to any course length and depth without losing its educational quality.

We provide a unifying framework where artificial neural networks and their architectures can be formally described as particular cases of a general mathematical construction---machines of finite depth. Unlike neural networks, machines have a precise definition, from which several properties follow naturally. Machines of finite depth are modular (they can be combined), efficiently computable, and differentiable. The backward pass of a machine is again a machine and can be computed without overhead using the same procedure as the forward pass. We prove this statement theoretically and practically via a unified implementation that generalizes several classical architectures---dense, convolutional, and recurrent neural networks with a rich shortcut structure---and their respective backpropagation rules.

Automatic diagnosis systems aim to probe for symptoms (i.e., symptom checking) and diagnose disease through multi-turn conversations with patients. Most previous works formulate it as a sequential decision process and use reinforcement learning (RL) to decide whether to inquire about symptoms or make a diagnosis. However, these RL-based methods heavily rely on the elaborate reward function and usually suffer from an unstable training process and low data efficiency. In this work, we propose an effective multi-task framework for automatic diagnosis called MTDiag. We first reformulate symptom checking as a multi-label classification task by direct supervision. Each medical dialogue is equivalently converted into multiple samples for classification, which can also help alleviate the data scarcity problem. Furthermore, we design a multi-task learning strategy to guide the symptom checking procedure with disease information and further utilize contrastive learning to better distinguish symptoms between diseases. Extensive experimental results show that our method achieves state-of-the-art performance on four public datasets with 1.7%~3.1% improvement in disease diagnosis, demonstrating the superiority of the proposed method. Additionally, our model is now deployed in an online medical consultant system as an assistant tool for real-life doctors.

Non-intrusive load monitoring (NILM) aims to decompose aggregated electrical usage signal into appliance-specific power consumption and it amounts to a classical example of blind source separation tasks. Leveraging recent progress on deep learning techniques, we design a new neural NILM model {\em Multi-State Dual CNN} (MSDC). Different from previous models, MSDC explicitly extracts information about the appliance's multiple states and state transitions, which in turn regulates the prediction of signals for appliances. More specifically, we employ a dual-CNN architecture: one CNN for outputting state distributions and the other for predicting the power of each state. A new technique is invented that utilizes conditional random fields (CRF) to capture state transitions. Experiments on two real-world datasets REDD and UK-DALE demonstrate that our model significantly outperform state-of-the-art models while having good generalization capacity, achieving 6%-10% MAE gain and 33%-51% SAE gain to unseen appliances.

Visible-infrared person re-identification (VI-ReID), which aims to search identities across different spectra, is a challenging task due to large cross-modality discrepancy between visible and infrared images. The key to reduce the discrepancy is to filter out identity-irrelevant interference and effectively learn modality-invariant person representations. In this paper, we propose a novel Modality Restitution and Compensation Network (MRCN) to narrow the gap between the two modalities. Specifically, we first reduce the modality discrepancy by using two Instance Normalization (IN) layers. Next, to reduce the influence of IN layers on removing discriminative information and to reduce modality differences, we propose a Modality Restitution Module (MRM) and a Modality Compensation Module (MCM) to respectively distill modality-irrelevant and modality-relevant features from the removed information. Then, the modality-irrelevant features are used to restitute to the normalized visible and infrared features, while the modality-relevant features are used to compensate for the features of the other modality. Furthermore, to better disentangle the modality-relevant features and the modality-irrelevant features, we propose a novel Center-Quadruplet Causal (CQC) loss to encourage the network to effectively learn the modality-relevant features and the modality-irrelevant features. Extensive experiments are conducted to validate the superiority of our method on the challenging SYSU-MM01 and RegDB datasets. More remarkably, our method achieves 95.1% in terms of Rank-1 and 89.2% in terms of mAP on the RegDB dataset.

Visual contents, such as illustrations and images, play a big role in product manual understanding. Existing Product Manual Question Answering (PMQA) datasets tend to ignore visual contents and only retain textual parts. In this work, to emphasize the importance of multimodal contents, we propose a Multimodal Product Manual Question Answering (MPMQA) task. For each question, MPMQA requires the model not only to process multimodal contents but also to provide multimodal answers. To support MPMQA, a large-scale dataset PM209 is constructed with human annotations, which contains 209 product manuals from 27 well-known consumer electronic brands. Human annotations include 6 types of semantic regions for manual contents and 22,021 pairs of question and answer. Especially, each answer consists of a textual sentence and related visual regions from manuals. Taking into account the length of product manuals and the fact that a question is always related to a small number of pages, MPMQA can be naturally split into two subtasks: retrieving most related pages and then generating multimodal answers. We further propose a unified model that can perform these two subtasks all together and achieve comparable performance with multiple task-specific models. The PM209 dataset is available at https://github.com/AIM3-RUC/MPMQA.

Multimodal named entity recognition (MNER) is a critical step in information extraction, which aims to detect entity spans and classify them to corresponding entity types given a sentence-image pair. Existing methods either (1) obtain named entities with coarse-grained visual clues from attention mechanisms, or (2) first detect fine-grained visual regions with toolkits and then recognize named entities. However, they suffer from improper alignment between entity types and visual regions or error propagation in the two-stage manner, which finally imports irrelevant visual information into texts. In this paper, we propose a novel end-to-end framework named MNER-QG that can simultaneously perform MRC-based multimodal named entity recognition and query grounding. Specifically, with the assistance of queries, MNER-QG can provide prior knowledge of entity types and visual regions, and further enhance representations of both text and image. To conduct the query grounding task, we provide manual annotations and weak supervisions that are obtained via training a highly flexible visual grounding model with transfer learning. We conduct extensive experiments on two public MNER datasets, Twitter2015 and Twitter2017. Experimental results show that MNER-QG outperforms the current state-of-the-art models on the MNER task, and also improves the query grounding performance.

Automatic medical report generation is an essential task in applying artificial intelligence to the medical domain, which can lighten the workloads of doctors and promote clinical automation. The state-of-the-art approaches employ Transformer-based encoder-decoder architectures to generate reports for medical images. However, they do not fully explore the relationships between multi-modal medical data, and generate inaccurate and inconsistent reports. To address these issues, this paper proposes a Multi-modal Memory Transformer Network (MMTN) to cope with multi-modal medical data for generating image-report consistent medical reports. On the one hand, MMTN reduces the occurrence of image-report inconsistencies by designing a unique encoder to associate and memorize the relationship between medical images and medical terminologies. On the other hand, MMTN utilizes the cross-modal complementarity of the medical vision and language for the word prediction, which further enhances the accuracy of generating medical reports. Extensive experiments on three real datasets show that MMTN achieves significant effectiveness over state-of-the-art approaches on both automatic metrics and human evaluation.