IJCAI 2020 Papers

ICD coding is a process of assigning the International Classification of Disease diagnosis codes to clinical/medical notes documented by health professionals (e.g. clinicians). This process requires significant human resources, and thus is costly and prone to error. To handle the problem, machine learning has been utilized for automatic ICD coding. Previous state-of-the-art models were based on convolutional neural networks, using a single/several fixed window sizes. However, the lengths and interdependence between text fragments related to ICD codes in clinical text vary significantly, leading to the difficulty of deciding what the best window sizes are. In this paper, we propose a new label attention model for automatic ICD coding, which can handle both the various lengths and the interdependence of the ICD code related text fragments. Furthermore, as the majority of ICD codes are not frequently used, leading to the extremely imbalanced data issue, we additionally propose a hierarchical joint learning mechanism extending our label attention model to handle the issue, using the hierarchical relationships among the codes. Our label attention model achieves new state-of-the-art results on three benchmark MIMIC datasets, and the joint learning mechanism helps improve the performances for infrequent codes.

We study complete approximations of an ontology formulated in a non-Horn description logic (DL) such as ALC in a Horn DL such as EL. We provide concrete approximation schemes that are necessarily infinite and observe that in the ELU-to-EL case finite approximations tend to exist in practice and are guaranteed to exist when the source ontology is acyclic. In contrast, neither of this is the case for ELU_bot-to-EL_bot and for ALC-to-EL_bot approximations. We also define a notion of approximation tailored towards ontology-mediated querying, connect it to subsumption-based approximations, and identify a case where finite approximations are guaranteed to exist.

The conventional single-target Cross-Domain Recommendation (CDR) only improves the recommendation accuracy on a target domain with the help of a source domain (with relatively richer information). In contrast, the novel dual-target CDR has been proposed to improve the recommendation accuracies on both domains simultaneously. However, dual-target CDR faces two new challenges: (1) how to generate more representative user and item embeddings, and (2) how to effectively optimize the user/item embeddings on each domain. To address these challenges, in this paper, we propose a graphical and attentional framework, called GA-DTCDR. In GA-DTCDR, we first construct two separate heterogeneous graphs based on the rating and content information from two domains to generate more representative user and item embeddings. Then, we propose an element-wise attention mechanism to effectively combine the embeddings of common users learned from both domains. Both steps significantly enhance the quality of user and item embeddings and thus improve the recommendation accuracy on each domain. Extensive experiments conducted on four real-world datasets demonstrate that GA-DTCDR significantly outperforms the state-of-the-art approaches.

Human-Object Interaction (HOI) detection devotes to learn how humans interact with surrounding objects via inferring triplets of < human, verb, object >. However, recent HOI detection methods mostly rely on additional annotations (e.g., human pose) and neglect powerful interactive reasoning beyond convolutions. In this paper, we present a novel graph-based interactive reasoning model called Interactive Graph (abbr. in-Graph) to infer HOIs, in which interactive semantics implied among visual targets are efficiently exploited. The proposed model consists of a project function that maps related targets from convolution space to a graph-based semantic space, a message passing process propagating semantics among all nodes and an update function transforming the reasoned nodes back to convolution space. Furthermore, we construct a new framework to assemble in-Graph models for detecting HOIs, namely in-GraphNet. Beyond inferring HOIs using instance features respectively, the framework dynamically parses pairwise interactive semantics among visual targets by integrating two-level in-Graphs, i.e., scene-wide and instance-wide in-Graphs. Our framework is end-to-end trainable and free from costly annotations like human pose. Extensive experiments show that our proposed framework outperforms existing HOI detection methods on both V-COCO and HICO-DET benchmarks and improves the baseline about 9.4% and 15% relatively, validating its efficacy in detecting HOIs.

Population aging is becoming an increasingly important issue around the world. As people live longer, they also tend to suffer from more challenging medical conditions. Currently, there is a lack of a holistic technology-powered solution for providing quality care at affordable cost to patients suffering from co-morbidity. In this paper, we demonstrate a novel AI-powered solution to provide early detection of the onset of Dementia + Parkinson's disease (DPD) co-morbidity, a condition which severely limits a senior's ability to live actively and independently. We investigate useful in-game behaviour markers which can support machine learning-based predictive analytics on seniors' risk of developing DPD co-morbidity.

K-cores are maximal induced subgraphs where all vertices have degree at least k. These dense patterns have applications in community detection, network visualization and protein function prediction. However, k-cores can be quite unstable to network modifications, which motivates the question: How resilient is the k-core structure of a network, such as the Web or Facebook, to edge deletions? We investigate this question from an algorithmic perspective. More specifically, we study the problem of computing a small set of edges for which the removal minimizes the k-core structure of a network. This paper provides a comprehensive characterization of the hardness of the k-core minimization problem (KCM), including innaproximability and parameterized complexity. Motivated by these challenges, we propose a novel algorithm inspired by Shapley value---a cooperative game-theoretic concept--- that is able to leverage the strong interdependencies in the effects of edge removals in the search space. We efficiently approximate Shapley values using a randomized algorithm with probabilistic guarantees. Our experiments, show that the proposed algorithm outperforms competing solutions in terms of k-core minimization while being able to handle large graphs. Moreover, we illustrate how KCM can be applied in the analysis of the k-core resilience of networks.

ASPIC+ is an established general framework for argumentation and non-monotonic reasoning. However, ASPIC+ does not satisfy the non-contamination rationality postulates, and moreover, tacitly assumes unbounded resources when demonstrating satisfaction of the consistency postulates. In this paper we present a new version of ASPIC+ – Dialectial ASPIC+ – that is fully rational under resource bounds.

Description logics are well-known logical formalisms for knowledge representation. We propose to enrich knowledge bases (KBs) with dynamic axioms that specify how the satisfaction of statements from the KBs evolves when the interpretation is decomposed or recomposed, providing a natural means to predict the evolution of interpretations. Our dynamic axioms borrow logical connectives from separation logics, well-known specification languages to verify programs with dynamic data structures. In the paper, we focus on ALC and EL augmented with dynamic axioms, or to their subclass of positive dynamic axioms. The knowledge base consistency problem in the presence of dynamic axioms is investigated, leading to interesting complexity results, among which the problem for EL with positive dynamic axioms is tractable, whereas EL with dynamic axioms is undecidable.

The issue of describing in a formal way solving algorithms in various fields such as Propositional Satisfiability (SAT), Quantified SAT, Satisfiability Modulo Theories, Answer Set Programming (ASP), and Constraint ASP, has been relatively recently solved employing abstract solvers. In this paper we deal with cautious reasoning tasks in ASP, and design, implement and test novel abstract solutions, borrowed from backbone computation in SAT. By employing abstract solvers, we also formally show that the algorithms for solving cautious reasoning tasks in ASP are strongly related to those for computing backbones of Boolean formulas. Some of the new solutions have been implemented in the ASP solver WASP, and tested.

Factorization Machines (FMs) refer to a class of general predictors working with real valued feature vectors, which are well-known for their ability to estimate model parameters under significant sparsity and have found successful applications in many areas such as the click-through rate (CTR) prediction. However, standard FMs only produce a single fixed representation for each feature across different input instances, which may limit the CTR model’s expressive and predictive power. Inspired by the success of Input-aware Factorization Machines (IFMs), which aim to learn more flexible and informative representations of a given feature according to different input instances, we propose a novel model named Dual Input-aware Factorization Machines (DIFMs) that can adaptively reweight the original feature representations at the bit-wise and vector-wise levels simultaneously. Furthermore, DIFMs strategically integrate various components including Multi-Head Self-Attention, Residual Networks and DNNs into a unified end-to-end model. Comprehensive experiments on two real-world CTR prediction datasets show that the DIFM model can outperform several state-of-the-art models consistently.

We present a novel negotiation model that allows an agent to learn how to negotiate during concurrent bilateral negotiations in unknown and dynamic e-markets. The agent uses an actor-critic architecture with model-free reinforcement learning to learn a strategy expressed as a deep neural network. We pre-train the strategy by supervision from synthetic market data, thereby decreasing the exploration time required for learning during negotiation. As a result, we can build automated agents for concurrent negotiations that can adapt to different e-market settings without the need to be pre-programmed. Our experimental evaluation shows that our deep reinforcement learning based agents outperform two existing well-known negotiation strategies in one-to-many concurrent bilateral negotiations for a range of e-market settings.

Due to the rising awareness of privacy protection and the voluminous scale of speech data, it is becoming infeasible for Automatic Speech Recognition (ASR) system developers to train the acoustic model with complete data as before. In this paper, we propose a novel Divide-and-Merge paradigm to solve salient problems plaguing the ASR field. In the Divide phase, multiple acoustic models are trained based upon different subsets of the complete speech data, while in the Merge phase two novel algorithms are utilized to generate a high-quality acoustic model based upon those trained on data subsets. We first propose the Genetic Merge Algorithm (GMA), which is a highly specialized algorithm for optimizing acoustic models but suffers from low efficiency. We further propose the SGD-Based Optimizational Merge Algorithm (SOMA), which effectively alleviates the efficiency bottleneck of GMA and maintains superior performance. Extensive experiments on public data show that the proposed methods can significantly outperform the state-of-the-art.

Analogical transfer consists in leveraging a measure of similarity between two situations to predict the amount of similarity between their outcomes. Acquiring a suitable similarity measure for analogical transfer may be difficult, especially when the data is sparse or when the domain knowledge is incomplete. To alleviate this problem, this paper presents a dataset complexity measure that can be used either to select an optimal similarity measure, or if the similarity measure is given, to perform analogical transfer: among the potential outcomes of a new situation, the most plausible is the one which minimizes the dataset complexity.

A generative probabilistic model for relational data consists of a family of probability distributions for relational structures over domains of different sizes. In most existing statistical relational learning (SRL) frameworks, these models are not projective in the sense that the marginal of the distribution for size-n structures on induced substructures of size k

Learning higher-level representations from data has been on the agenda of AI research for several decades. In the paper, I will give a survey of various approaches to learning symbolic higher-level representations: feature construction and constructive induction, predicate invention, propositionalization, pattern mining, and mining time series patterns. Finally, I will give an outlook on how approaches to learning higher-level representations, symbolic and neural, can benefit from each other to solve current issues in machine learning.

Practically, we often face the dilemma that some of the examples for training a classifier are incorrectly labeled due to various subjective and objective factors. Although intensive efforts have been put to design classifiers that are robust to label noise, most of the previous methods have not fully utilized data distribution information. To address this issue, this paper introduces a bi-level learning paradigm termed “Spectral Cluster Discovery'' (SCD) for combating with noisy labels. Namely, we simultaneously learn a robust classifier (Learning stage) by discovering the low-rank approximation to the ground-truth label matrix and learn an ideal affinity graph (Clustering stage). Specifically, we use the learned classifier to assign the examples with similar label to a mutual cluster. Based on the cluster membership, we utilize the learned affinity graph to explore the noisy examples based on the cluster membership. Both stages will reinforce each other iteratively. Experimental results on typical benchmark and real-world datasets verify the superiority of SCD to other label noise learning methods.

We formulate object segmentation in video as a spectral graph clustering problem in space and time, in which nodes are pixels and their relations form local neighbourhoods. We claim that the strongest cluster in this pixel-level graph represents the salient object segmentation. We compute the main cluster using a novel and fast 3D filtering technique that finds the spectral clustering solution, namely the principal eigenvector of the graph's adjacency matrix, without building the matrix explicitly - which would be intractable. Our method is based on the power iteration which we prove is equivalent to performing a specific set of 3D convolutions in the space-time feature volume. This allows us to avoid creating the matrix and have a fast parallel implementation on GPU. We show that our method is much faster than classical power iteration applied directly on the adjacency matrix. Different from other works, ours is dedicated to preserving object consistency in space and time at the level of pixels. In experiments, we obtain consistent improvement over the top state of the art methods on DAVIS-2016 dataset. We also achieve top results on the well-known SegTrackv2 dataset.