IJCAI 2023 论文列表

This paper proposes BPNet, a novel end-to-end deep learning framework to learn Bézier primitive segmentation on 3D point clouds. The existing works treat different primitive types separately, thus limiting them to finite shape categories. To address this issue, we seek a generalized primitive segmentation on point clouds. Taking inspiration from Bézier decomposition on NURBS models, we transfer it to guide point cloud segmentation casting off primitive types. A joint optimization framework is proposed to learn Bézier primitive segmentation and geometric fitting simultaneously on a cascaded architecture. Specifically, we introduce a soft voting regularizer to improve primitive segmentation and propose an auto-weight embedding module to cluster point features, making the network more robust and generic. We also introduce a reconstruction module where we successfully process multiple CAD models with different primitives simultaneously. We conducted extensive experiments on the synthetic ABC dataset and real-scan datasets to validate and compare our approach with different baseline methods. Experiments show superior performance over previous work in terms of segmentation, with a substantially faster inference speed.

Federated learning (FL) is a prospective distributed machine learning framework that can preserve data privacy. In particular, cross-silo FL can complete model training by making isolated data islands of different organizations collaborate with a parameter server (PS) via exchanging model parameters for multiple communication rounds. In cross-silo FL, an incentive mechanism is indispensable for motivating data owners to contribute their models to FL training. However, how to allocate the reward budget among different rounds is an essential but complicated problem largely overlooked by existing works. The challenge of this problem lies in the opaque feedback between reward budget allocation and model utility improvement of FL, making the optimal reward budget allocation complicated. To address this problem, we design an online reward budget allocation algorithm using Bayesian optimization named BARA (Budget Allocation for Reverse Auction). Specifically, BARA can model the complicated relationship between reward budget allocation and final model accuracy in FL based on historical training records so that the reward budget allocated to each communication round is dynamically optimized so as to maximize the final model utility. We further incorporate the BARA algorithm into reverse auction-based incentive mechanisms to illustrate its effectiveness. Extensive experiments are conducted on real datasets to demonstrate that BARA significantly outperforms competitive baselines by improving model utility with the same amount of reward budget.

We consider the problem of navigating in a Markov decision process where extrinsic rewards are either absent or ignored. In this setting, the objective is to learn policies to reach all the states that are reachable within a given number of steps (in expectation) from a starting state. We introduce a novel meta-algorithm which can use any online reinforcement learning algorithm (with appropriate regret guarantees) as a black-box. Our algorithm demonstrates a method for transforming the output of online algorithms to a batch setting. We prove an upper bound on the sample complexity of our algorithm in terms of the regret bound of the used black-box RL algorithm. Furthermore, we provide experimental results to validate the effectiveness of our algorithm and correctness of our theoretical results.

Rigid origami has shown potential in large diversity of practical applications. However, current rigid origami crease pattern design mostly relies on known tessellations. This strongly limits the diversity and novelty of patterns that can be created. In this work, we build upon the recently developed principle of three units method to formulate rigid origami design as a discrete optimization problem, the rigid origami game. Our implementation allows for a simple definition of diverse objectives and thereby expands the potential of rigid origami further to optimized, application-specific crease patterns. We showcase the flexibility of our formulation through use of a diverse set of search methods in several illustrative case studies. We are not only able to construct various patterns that approximate given target shapes, but to also specify abstract, function-based rewards which result in novel, foldable and functional designs for everyday objects.

Generalized planning (GP) studies the computation of general solutions for a set of planning problems. Computing general solutions with correctness guarantee has long been a key issue in GP. Abstractions are widely used to solve GP problems. For example, a popular abstraction model for GP is qualitative numeric planning (QNP), which extends classical planning with non-negative real variables that can be increased or decreased by some arbitrary amount. The refinement of correct solutions of sound abstractions are solutions with correctness guarantees for GP problems. More recent literature proposed a uniform abstraction framework for GP and gave model-theoretic definitions of sound and complete abstractions for GP problems. In this paper, based on the previous work, we explore automatic verification of sound abstractions for GP. Firstly, we present a proof-theoretic characterization for sound abstractions. Secondly, based on the characterization, we give a sufficient condition for sound abstractions with deterministic actions. Then we study how to verify the sufficient condition when the abstraction models are bounded QNPs where integer variables can be incremented or decremented by one. To this end, we develop methods to handle counting and transitive closure, which are often used to define numerical variables. Finally, we implement a sound bounded QNP abstraction verification system and report experimental results on several domains.

Truss layout design, namely finding a lightweight truss layout satisfying all the physical constraints, is a fundamental problem in the building industry. Generating the optimal layout is a challenging combinatorial optimization problem, which can be extremely expensive to solve by exhaustive search. Directly applying end-to-end reinforcement learning (RL) methods to truss layout design is infeasible either, since only a tiny portion of the entire layout space is valid under the physical constraints, leading to particularly sparse rewards for RL training. In this paper, we develop AutoTruss, a two-stage framework to efficiently generate both lightweight and valid truss layouts. AutoTruss first adopts Monte Carlo tree search to discover a diverse collection of valid layouts. Then RL is applied to iteratively refine the valid solutions. We conduct experiments and ablation studies in popular truss layout design test cases in both 2D and 3D settings. AutoTruss outperforms the best-reported layouts by 25.1% in the most challenging 3D test cases, resulting in the first effective deep-RL-based approach in the truss layout design literature.

From the perspective of a dialog system, the identification of the intention behind the segments in a dialog is important, as it provides cues regarding the information present in the segments and how they should be interpreted. The ISO 24617-2 standard for dialog act annotation defines a hierarchically organized set of general-purpose communicative functions that correspond to different intentions that are relevant in the context of a dialog. In this paper, we explore the automatic recognition of these functions. To do so, we propose to adapt existing approaches to dialog act recognition, so that they can deal with the hierarchical classification problem. More specifically, we propose the use of an end-to-end hierarchical network with cascading outputs and maximum a posteriori path estimation to predict the communicative function at each level of the hierarchy, preserve the dependencies between the functions in the path, and decide at which level to stop. Additionally, we rely on transfer learning processes to address the data scarcity problem. Our experiments on the DialogBank show that this approach outperforms both flat and hierarchical approaches based on multiple classifiers and that each of its components plays an important role in the recognition of general-purpose communicative functions.

There is a growing interest in the use of AI techniques for urban traffic control, with a particular focus on traffic signal optimisation. Model-based approaches such as planning demonstrated to be capable of dealing in real-time with unexpected or unusual traffic conditions, as well as with the usual traffic patterns. Further, the knowledge models on which such techniques rely to generate traffic signal strategies are in fact simulation models of traffic, hence can be used by traffic authorities to test and compare different approaches. In this work, we present a framework that relies on automated planning to generate and simulate traffic signal strategies in a urban region. To demonstrate the capabilities of the framework, we consider real-world data collected from sensors deployed in a major corridor of the Kirklees region of the United Kingdom.

Since the dawn of Transformer-based models, the trade-off between transparency and accuracy has been a topical issue in the NLP community. Working towards ethical and transparent automated content moderation (ACM), my goal is to find where it is still relevant to implement linguistic expertise. I show that transparent statistical models based on linguistic knowledge can still be competitive, while linguistic features have many other useful applications.

Automated machine learning (AutoML) has been widely researched and adopted for supervised problems, but progress in unsupervised settings has been limited. We propose `"LOTUS", a novel framework to automate outlier detection based on meta-learning. Our premise is that the selection of the optimal outlier detection technique depends on the inherent properties of the data distribution. We leverage optimal transport to find the dataset with the most similar underlying distribution, and then apply the outlier detection techniques that proved to work best for that data distribution. We evaluate the robustness of our framework and find that it outperforms all state-of-the-art automated outlier detection tools. This approach can also be easily generalized to automate other unsupervised settings.

In online advertising markets, an increasing number of advertisers are adopting auto-bidders to buy advertising slots. This tool simplifies the process of optimizing bids based on various financial constraints. In our study, we focus on second-price auctions where bidders have both private budget and private ROI (return on investment) constraints. We formulate the auto-bidding system design problem as a mathematical program and analyze the auto-bidders' bidding strategy under such constraints. We demonstrate that our design ensures truthfulness, i.e., among all pure and mixed strategies, always reporting the truthful budget and ROI is an optimal strategy for the bidders. Although the program is non-convex, we provide a fast algorithm to compute the optimal bidding strategy for the bidders based on our analysis. We also study the welfare and provide a lower bound for the PoA (price of anarchy). Moreover, we prove that if all bidders utilize our auto-bidding system, a Bayesian Nash equilibrium exists. We provide a sufficient condition under which the iterated best response process converges to such an equilibrium. Finally, we conduct extensive experiments to empirically evaluate the effectiveness of our design.

Spectrum-based Fault Localization (SBFL) uses the coverage of test cases and their outcome (pass/fail) to predict the "suspiciousness'' of program components, e.g., lines of code. SBFL is, perhaps, the most successful fault localization technique due to its simplicity and scalability. However, SBFL heuristics do not perform well in scenarios where a program may have multiple faulty components. In this work, we propose a new algorithm that "augments'' previously proposed SBFL heuristics to produce a ranked list where faulty components ranked low by base SBFL metrics are ranked significantly higher. We implement our ideas in a tool, ARTEMIS, that attempts to "bubble up'' faulty components which are ranked lower by base SBFL metrics. We compare our technique to the most popular SBFL metrics and demonstrate statistically significant improvement in the developer effort for fault localization with respect to the basic strategies.

Image-based quick response (QR) code is frequently used, but creates barriers for the visual impaired people. With the goal of ``AI for good", this paper proposes the AudioQR, a barrier-free QR coding mechanism for the visually impaired population via deep neural audio watermarks. Previous audio watermarking approaches are mainly based on handcrafted pipelines, which is less secure and difficult to apply in large-scale scenarios. In contrast, AudioQR is the first comprehensive end-to-end pipeline that hides watermarks in audio imperceptibly and robustly. To achieve this, we jointly train an encoder and decoder, where the encoder is structured as a concatenation of transposed convolutions and multi-receptive field fusion modules. Moreover, we customize the decoder training with a stochastic data augmentation chain to make the watermarked audio robust towards different audio distortions, such as environment background, room impulse response when playing through the air, music surrounding, and Gaussian noise. Experiment results indicate that AudioQR can efficiently hide arbitrary information into audio without introducing significant perceptible difference. Our code is available at https://github.com/xinghua-qu/AudioQR.

Multi-agent task allocation in physical environments with spatial and temporal constraints, are hard problems that are relevant in many realistic applications. A task allocation algorithm based on Fisher market clearing (FMC_TA), that can be performed either centrally or distributively, has been shown to produce high quality allocations in comparison to both centralized and distributed state of the art incomplete optimization algorithms. However, the algorithm is synchronous and therefore depends on perfect communication between agents. We propose FMC_ATA, an asynchronous version of FMC_TA, which is robust to message latency and message loss. In contrast to the former version of the algorithm, FMC_ATA allows agents to identify dynamic events and initiate the generation of an updated allocation. Thus, it is more compatible for dynamic environments. We further investigate the conditions in which the distributed version of the algorithm is preferred over the centralized version. Our results indicate that the proposed asynchronous distributed algorithm produces consistent results even when the communication level is extremely poor.

A significant challenge in detecting and mitigating bias is creating a mindset amongst AI developers to address unfairness. The current literature on fairness is broad, and the learning curve to distinguish where to use existing metrics and techniques for bias detection or mitigation is difficult. This survey systematises the state-of-the-art about distinct notions of fairness and relative techniques for bias mitigation according to the AI lifecycle. Gaps and challenges identified during the development of this work are also discussed.

Although ChatGPT attempts to mitigate bias, when instructed to translate the gender-neutral Turkish sentences “O bir doktor. O bir hemşire” to English, the outcome is biased: “He is a doctor. She is a nurse.” In 2016, we have demonstrated that language representations trained via unsupervised learning automatically embed implicit biases documented in social cognition through the statistical regularities in language corpora. Evaluating embedding associations in language, vision, and multi-modal language-vision models reveals that large-scale sociocultural data is a source of implicit human biases regarding gender, race or ethnicity, skin color, ability, age, sexuality, religion, social class, and intersectional associations. The study of gender bias in language, vision, language-vision, and generative AI has highlighted the sexualization of women and girls in AI, while easily accessible generative AI models such as text-to-image generators amplify bias at scale. As AI increasingly automates tasks that determine life’s outcomes and opportunities, the ethics of AI bias has significant implications for human cognition, society, justice, and the future of AI. Thus, it is necessary to advance our understanding of the depth, prevalence, and complexities of bias in AI to mitigate it both in machines and society.

Multi-agent literature explores personifying artificial agents with personality, emotions or cognitive biases to produce “typical”, believable agents. In this study, we demonstrate the potential of endowing artificial agents with a motivation, using human implicit motivation psychology theory that introduces 3 motive profiles - power, achievement and affiliation, to create diverse, risk-aware agents. We first devise a framework to model these motivated agents (or agents with any inherent behavior), that can activate different strategies depending on the circumstances. We conduct experiments on a fire-fighting task domain, evaluate how motivated teams perform, and draw conclusions on appropriate team compositions to be deployed in environments with different risk levels. Our framework generates predictable agents as their resulting behaviors align with the inherent characteristics of their motives. We find that motivational diversity within teams is beneficial in dynamic collaborative environments, especially as the task risk level increases. Furthermore, we observed that the best composition in terms of the performance metrics used to evaluate team compositions, does not remain the same as the collaboration level required to achieve goals changes. These results have implications for future designs of risk-aware autonomous teams and Human-AI teams, as they highlight the prospects of creating better artificial teammates and performance gains that could be achieved through anthropomorphized motivated agents.

Causal reasoning reflects how humans perceive events in the world and establish relationships among them, identifying some as causes and others as effects. Causal discovery is about agreeing on these relationships and drawing them as a causal graph. Argumentation is the way humans reason systematically about an idea: the medium we use to exchange opinions, to get to know and trust each other and possibly agree on controversial matters. Developing AI which can argue with humans about causality would allow us to understand and validate the analysis of the AI and would allow the AI to bring evidence for or against humans' prior knowledge. This is the goal of this project: to develop a novel scientific paradigm of interactive causal discovery and train AI to recognise causes and effects by debating, with humans, the results of different statistical methods

We study the problem of fair allocation of indivisible goods that form a graph and the bundles that are distributed to agents are connected subgraphs of this graph. We focus on the maximin share and the proportional fairness criteria. It is well-known that allocations satisfying these criteria may not exist for many graphs including complete graphs and cycles. Therefore, it is natural to look for approximate allocations, i.e., allocations guaranteeing each agent a certain portion of the value that is satisfactory to her. In this paper we consider the class of graphs of goods which do not contain a star with d+1 edges (where d > 1) as an induced subgraph. For this class of graphs we prove that there is an allocation assigning each agent a connected bundle of value at least 1/d of her maximin share. Moreover, for the same class of graphs of goods, we show a theorem which specifies what fraction of the proportional share can be guaranteed to each agent if the values of single goods for the agents are bounded by a given fraction of this share.