IJCAI 2022 论文列表

In this paper, we present a novel double diffusion based neural radiance field, dubbed DD-NeRF, to reconstruct human body geometry and render the human body appearance in novel views from a sparse set of images. We first propose a double diffusion mechanism to achieve expressive representations of input images by fully exploiting human body priors and image appearance details at two levels. At the coarse level, we first model the coarse human body poses and shapes via an unclothed 3D deformable vertex model as guidance. At the fine level, we present a multi-view sampling network to capture subtle geometric deformations and image detailed appearances, such as clothing and hair, from multiple input views. Considering the sparsity of the two level features, we diffuse them into feature volumes in the canonical space to construct neural radiance fields. Then, we present a signed distance function (SDF) regression network to construct body surfaces from the diffused features. Thanks to our double diffused representations, our method can even synthesize novel views of unseen subjects. Experiments on various datasets demonstrate that our approach outperforms the state-of-the-art in both geometric reconstruction and novel view synthesis.

Learning complex programs through inductive logic programming (ILP) remains a formidable challenge. Existing higher-order enabled ILP systems show improved accuracy and learning performance, though remain hampered by the limitations of the underlying learning mechanism. Experimental results show that our extension of the versatile Learning From Failures paradigm by higher-order definitions significantly improves learning performance without the burdensome human guidance required by existing systems. Our theoretical framework captures a class of higher-order definitions preserving soundness of existing subsumption-based pruning methods.

Group activity recognition aims to understand the overall behavior performed by a group of people. Recently, some graph-based methods have made progress by learning the relation graphs among multiple persons. However, the differences between an individual and others play an important role in identifying confusable group activities, which have not been elaborately explored by previous methods. In this paper, a novel Graph-based Residual AggregatIon Network (GRAIN) is proposed to model the differences among all persons of the whole group, which is end-to-end trainable. Specifically, a new local residual relation module is explicitly proposed to capture the local spatiotemporal differences of relevant persons, which is further combined with the multi-graph relation networks. Moreover, a weighted aggregation strategy is devised to adaptively select multi-level spatiotemporal features from the appearance-level information to high level relations. Finally, our model is capable of extracting a comprehensive representation and inferring the group activity in an end-to-end manner. The experimental results on two popular benchmarks for group activity recognition clearly demonstrate the superior performance of our method in comparison with the state-of-the-art methods.

Gaussian mixture model (GMM) is a powerful statistical tool in data modeling, especially for unsupervised learning tasks. Traditional learning methods for GMM such as expectation maximization (EM) require the covariance of the Gaussian components to be non-singular, a condition that is often not satisfied in real-world applications. This paper presents a new learning method called G$^2$M$^2$ (General Gaussian Mixture Model) by fitting an unnormalized Gaussian mixture function (UGMF) to a data distribution. At the core of G$^2$M$^2$ is the introduction of an integral cosine similarity (ICS) function for comparing the UGMF and the unknown data density distribution without having to explicitly estimate it. By maximizing the ICS through Monte Carlo sampling, the UGMF can be made to overlap with the unknown data density distribution such that the two only differ by a constant scalar, and the UGMF can be normalized to obtain the data density distribution. A Siamese convolutional neural network is also designed for optimizing the ICS function. Experimental results show that our method is more competitive in modeling data having correlations that may lead to singular covariance matrices in GMM, and it outperforms state-of-the-art methods in unsupervised anomaly detection.

Learning first-order logic programs (LPs) from relational facts which yields intuitive insights into the data is a challenging topic in neuro-symbolic research. We introduce a novel differentiable inductive logic programming (ILP) model, called differentiable first-order rule learner (DFOL), which finds the correct LPs from relational facts by searching for the interpretable matrix representations of LPs. These interpretable matrices are deemed as trainable tensors in neural networks (NNs). The NNs are devised according to the differentiable semantics of LPs. Specifically, we first adopt a novel propositionalization method that transfers facts to NN-readable vector pairs representing interpretation pairs. We replace the immediate consequence operator with NN constraint functions consisting of algebraic operations and a sigmoid-like activation function. We map the symbolic forward-chained format of LPs into NN constraint functions consisting of operations between subsymbolic vector representations of atoms. By applying gradient descent, the trained well parameters of NNs can be decoded into precise symbolic LPs in forward-chained logic format. We demonstrate that DFOL can perform on several standard ILP datasets, knowledge bases, and probabilistic relation facts and outperform several well-known differentiable ILP models. Experimental results indicate that DFOL is a precise, robust, scalable, and computationally cheap differentiable ILP model.

Dense Retrieval (DR) has achieved state-of-the-art first-stage ranking effectiveness. However, the efficiency of most existing DR models is limited by the large memory cost of storing dense vectors and the time-consuming nearest neighbor search (NNS) in vector space. Therefore, we present RepCONC, a novel retrieval model that learns discrete Representations via CONstrained Clustering. RepCONC jointly trains dual-encoders and the Product Quantization (PQ) method to learn discrete document representations and enables fast approximate NNS with compact indexes. It models quantization as a constrained clustering process, which requires the document embeddings to be uniformly clustered around the quantization centroids. We theoretically demonstrate the importance of the uniform clustering constraint and derive an efficient approximate solution for constrained clustering by reducing it to an instance of the optimal transport problem. Extensive experiments on two popular ad-hoc retrieval benchmarks show that RepCONC substantially outperforms a wide range of existing retrieval models in terms of retrieval effectiveness, memory efficiency, and time efficiency.

Acquiring degraded images with paired high-resolution (HR) images is often challenging, impeding the advance of image super-resolution in real-world applications. By generating realistic low-resolution (LR) images with degradation similar to that in real-world scenarios, simulated paired LR-HR data can be constructed for supervised training. However, most of the existing work ignores the degradation uncertainty of the generated realistic LR images, since only one LR image has been generated given an HR image. To address this weakness, we propose learning the degradation uncertainty of generated LR images and sampling multiple LR images from the learned LR image (mean) and degradation uncertainty (variance) and construct LR-HR pairs to train the super-resolution (SR) networks. Specifically, uncertainty can be learned by minimizing the proposed loss based on Kullback-Leibler (KL) divergence. Furthermore, the uncertainty in the feature domain is exploited by a novel perceptual loss; and we propose to calculate the adversarial loss from the gradient information in the SR stage for stable training performance and better visual quality. Experimental results on popular real-world datasets show that our proposed method has performed better than other unsupervised approaches.

The combination of tree search and neural networks has achieved super-human performance in challenging domains. We are interested in transferring to humans the knowledge these learning systems generate. We hypothesize the process in which neural-guided tree search algorithms learn how to solve a set of problems can be used to generate curricula for helping human learners. In this paper we show how the Bootstrap learning system can be modified to learn curricula for humans in a puzzle domain. We evaluate our system in two curriculum learning settings. First, given a small set of problem instances, our system orders the instances to ease the learning process of human learners. Second, given a large set of problem instances, our system returns a small ordered subset of the initial set that can be presented to human learners. We evaluate our curricula with a user study where participants learn how to solve a class of puzzles from the game `The Witness.' The user-study results suggest one of the curricula our system generates compares favorably with simple baselines and is competitive with the curriculum from the original `The Witness' game in terms of user retention and effort.

Learning graph structure for graph neural networks (GNNs) is crucial to facilitate the GNN-based downstream learning tasks. It is challenging due to the non-differentiable discrete graph structure and lack of ground-truth. In this paper, we address these problems and propose a novel graph structure learning framework for GNNs. Firstly, we directly model the continuous graph structure with dual-normalization, which implicitly imposes sparse constraint and reduces the influence of noisy edges. Secondly, we formulate the whole training process as a bilevel programming problem, where the inner objective is to optimize the GNNs given learned graphs, while the outer objective is to optimize the graph structure to minimize the generalization error of downstream task. Moreover, for bilevel optimization, we propose an improved Neumann-IFT algorithm to obtain an approximate solution, which is more stable and accurate than existing optimization methods. Besides, it makes the bilevel optimization process memory-efficient and scalable to large graphs. Experiments on node classification and scene graph generation show that our method can outperform related methods, especially with noisy graphs.

An adversary can fool deep neural network object detectors by generating adversarial noises. Most of the existing works focus on learning local visible noises in an adversarial "patch" fashion. However, the 2D patch attached to a 3D object tends to suffer from an inevitable reduction in attack performance as the viewpoint changes. To remedy this issue, this work proposes the Coated Adversarial Camouflage (CAC) to attack the detectors in arbitrary viewpoints. Unlike the patch trained in the 2D space, our camouflage generated by a conceptually different training framework consists of 3D rendering and dense proposals attack. Specifically, we make the camouflage perform 3D spatial transformations according to the pose changes of the object. Based on the multi-view rendering results, the top-n proposals of the region proposal network are fixed, and all the classifications in the fixed dense proposals are attacked simultaneously to output errors. In addition, we build a virtual 3D scene to fairly and reproducibly evaluate different attacks. Extensive experiments demonstrate the superiority of CAC over the existing attacks, and it shows impressive performance both in the virtual scene and the real world. This poses a potential threat to the security-critical computer vision systems.

Many real-world phenomena arise from causal relationships among a set of variables. As a powerful tool, Bayesian Network (BN) has been successful in describing high-dimensional distributions. However, the faithfulness condition, enforced in most BN learning algorithms, is violated in the settings where multiple variables synergistically affect the outcome (i.e., with polyadic dependencies). Building upon recent development in cluster causal diagrams (C-DAGs), we initiate the formal study of learning C-DAGs from observational data to relax the faithfulness condition. We propose a new scoring function, the Clustering Information Criterion (CIC), based on information-theoretic measures that represent various complex interactions among variables. The CIC score also contains a penalization of the model complexity under the minimum description length principle. We further provide a searching strategy to learn structures of high scores. Experiments on both synthetic and real data support the effectiveness of the proposed method.

Good training data is a prerequisite to develop useful Machine Learning applications. However, in many domains existing data sets cannot be shared due to privacy regulations (e.g., from medical studies). This work investigates a simple yet unconventional approach for anonymized data synthesis to enable third parties to benefit from such anonymized data. We explore the feasibility of learning implicitly from visually unrealistic, task-relevant stimuli, which are synthesized by exciting the neurons of a trained deep neural network. As such, neuronal excitation can be used to generate synthetic stimuli. The stimuli data is used to train new classification models. Furthermore, we extend this framework to inhibit representations that are associated with specific individuals. Extensive comparative empirical investigation shows that different algorithms trained on the stimuli are able to generalize successfully on the same task as the original model.

Modeling the spread of social contagions is central to various applications in social computing. In this paper, we study the learnability of the competitive threshold model from a theoretical perspective. We demonstrate how competitive threshold models can be seamlessly simulated by artificial neural networks with finite VC dimensions, which enables analytical sample complexity and generalization bounds. Based on the proposed hypothesis space, we design efficient algorithms under the empirical risk minimization scheme. The theoretical insights are finally translated into practical and explainable modeling methods, the effectiveness of which is verified through a sanity check over a few synthetic and real datasets. The experimental results promisingly show that our method enjoys a decent performance without using excessive data points, outperforming off-the-shelf methods.

With the privatization deployment of DNNs on edge devices, the security of on-device DNNs has raised significant concern. To quantify the model leakage risk of on-device DNNs automatically, we propose NNReverse, the first learning-based method which can reverse DNNs from AI programs without domain knowledge. NNReverse trains a representation model to represent the semantics of binary code for DNN layers. By searching the most similar function in our database, NNReverse infers the layer type of a given function’s binary code. To represent assembly instructions semantics precisely, NNReverse proposes a more fine-grained embedding model to represent the textual and structural-semantic of assembly functions.

Optimal execution is a sequential decision-making problem for cost-saving in algorithmic trading. Studies have found that reinforcement learning (RL) can help decide the order-splitting sizes. However, a problem remains unsolved: how to place limit orders at appropriate limit prices? The key challenge lies in the ``continuous-discrete duality'' of the action space. On the one hand, the continuous action space using percentage changes in prices is preferred for generalization. On the other hand, the trader eventually needs to choose limit prices discretely due to the existence of the tick size, which requires specialization for every single stock with different characteristics (e.g., the liquidity and the price range). So we need continuous control for generalization and discrete control for specialization. To this end, we propose a hybrid RL method to combine the advantages of both of them. We first use a continuous control agent to scope an action subset, then deploy a fine-grained agent to choose a specific limit price. Extensive experiments show that our method has higher sample efficiency and better training stability than existing RL algorithms and significantly outperforms previous learning-based methods for order execution.

Large Neighbourhood Search (LNS) is an algorithmic framework for optimization problems that can yield good performance in many domains. In this paper, we present a method for applying LNS to improve anytime maximum satisfiability (MaxSAT) solving by introducing a neighbourhood selection policy that shows good empirical performance. We show that our LNS solver can often improve the suboptimal solutions produced by other anytime MaxSAT solvers. When starting with a suboptimal solution of reasonable quality, our approach often finds a better solution than the original anytime solver can achieve. We demonstrate that implementing our LNS solver on top of three different state-of-the-art anytime solvers improves the anytime performance of all three solvers within the standard time limit used in the incomplete tracks of the annual MaxSAT Evaluation.

Local search is a popular technique to solve combinatorial optimization problems efficiently. To escape local minima one generally uses metaheuristics or try to design large neighborhoods around the current best solution. A somewhat more black box approach consists in using an optimization solver to explore a large neighborhood. This is the large-neighborhood search (LNS) idea that we reuse in this work. We introduce a generic neighborhood exploration algorithm based on restricted decision diagrams (DD) constructed from the current best solution. We experiment DD-LNS on two sequencing problems: the traveling salesman problem with time windows (TSPTW) and a production planning problem (DLSP). Despite its simplicity, DD-LNS is competitive with the state-of-the-art MIP approach on DLSP. It is able to improve the best known solutions of some standard instances for TSPTW and even to prove the optimality of quite a few other instances.

Knowledge embeddings (KE) represent a knowledge graph (KG) by embedding entities and relations into continuous vector spaces. Existing methods are mainly structure-based or description-based. Structure-based methods learn representations that preserve the inherent structure of KGs. They cannot well represent abundant long-tail entities in real-world KGs with limited structural information. Description-based methods leverage textual information and language models. Prior approaches in this direction barely outperform structure-based ones, and suffer from problems like expensive negative sampling and restrictive description demand. In this paper, we propose LMKE, which adopts Language Models to derive Knowledge Embeddings, aiming at both enriching representations of long-tail entities and solving problems of prior description-based methods. We formulate description-based KE learning with a contrastive learning framework to improve efficiency in training and evaluation. Experimental results show that LMKE achieves state-of-the-art performance on KE benchmarks of link prediction and triple classification, especially for long-tail entities.

While state-of-the-art planning systems need a grounded (propositional) task representation, the input model is provided "lifted", specifying predicates and action schemas with variables over a finite object universe. The size of the grounded model is exponential in predicate/action-schema arity, limiting applicability to cases where it is small enough. Recent work has taken up this challenge, devising an effective lifted forward search planner as basis for lifted heuristic search, as well as a variety of lifted heuristic functions based on the delete relaxation. Here we add a novel family of lifted heuristic functions, based on landmarks. We design two methods for landmark extraction in the lifted setting. The resulting heuristics exhibit performance advantages over previous heuristics in several benchmark domains. Especially the combination with lifted delete relaxation heuristics to a LAMA-style planner yields good results, beating the previous state of the art in lifted planning.