Accurate graph classification via two-staged contrastive curriculum learning.

Given a graph dataset, how can we generate meaningful graph representations that maximize classification accuracy? Learning representative graph embeddings is important for solving various real-world graph-based tasks. Graph contrastive learning aims to learn representations of graphs by capturing the relationship between the original graph and the augmented graph. However, previous contrastive learning methods neither capture semantic information within graphs nor consider both nodes and graphs while learning graph embeddings. We propose TAG (Two-staged contrAstive curriculum learning for Graphs), a two-staged contrastive learning method for graph classification. TAG learns graph representations in two levels: node-level and graph level, by exploiting six degree-based model-agnostic augmentation algorithms. Experiments show that TAG outperforms both unsupervised and supervised methods in classification accuracy, achieving up to 4.08% points and 4.76% points higher than the second-best unsupervised and supervised methods on average, respectively.

PET: Parameter-efficient Knowledge Distillation on Transformer.

Given a large Transformer model, how can we obtain a small and computationally efficient model which maintains the performance of the original model? Transformer has shown significant performance improvements for many NLP tasks in recent years. However, their large size, expensive computational cost, and long inference time make it challenging to deploy them to resource-constrained devices. Existing Transformer compression methods mainly focus on reducing the size of the encoder ignoring the fact that the decoder takes the major portion of the long inference time. In this paper, we propose PET (Parameter-Efficient knowledge distillation on Transformer), an efficient Transformer compression method that reduces the size of both the encoder and decoder. In PET, we identify and exploit pairs of parameter groups for efficient weight sharing, and employ a warm-up process using a simplified task to increase the gain through Knowledge Distillation. Extensive experiments on five real-world datasets show that PET outperforms existing methods in machine translation tasks. Specifically, on the IWSLT'14 EN→DE task, PET reduces the memory usage by 81.20% and accelerates the inference speed by 45.15% compared to the uncompressed model, with a minor decrease in BLEU score of 0.27.

Endoscope-assisted microsurgical clipping and reduction of post-clipping cerebral infarction: historical comparison using diffusion-weighted images.

Endoscopic assistance for aneurysm clipping and its possible benefits have been suggested in previous studies, but its clinical significance has not been fully elucidated. This study aimed to present the efficacy of endoscopy-assisted clipping in reducing post-clipping cerebral infarction (PCI) and clinical outcomes via a historical comparison of patients in our institution from January 2020 to March 2022. A total of 348 patients were included, 189 of whom underwent endoscope-assisted clipping. The overall incidence of PCI was 10.9% (n = 38); it was 15.7% (n = 25) before applying endoscopic assistance and decreased to 6.9% (n = 13) after endoscope application (p = 0.010). The application of a temporary clip (odds ratio [OR]: 2.673, 95% confidence interval [CI]: 1.291-5.536), history of hypertension (OR: 2.176, 95% CI: 0.897-5.279), history of diabetes mellitus (OR: 2.530, 95% CI: 1.079-5.932), and current smoker (OR: 3.553, 95% CI: 1.288-9.802) were independent risk factors of PCI, whereas endoscopic assistance was an independent inverse risk factor (OR: 0.387, 95% CI: 0.182-0.823). Compared to the location of the unruptured intracranial aneurysms, internal carotid artery aneurysms showed a significant decrease in the incidence of PCI (5.8% vs. 22.9%, p = 0.019). In terms of clinical outcomes, PCI was a significant risk factor for longer admission duration, intensive care unit stay, and poor clinical outcomes. However, endoscopic assistance itself was not a significant risk factor for clinical outcomes on the 45-day modified Rankin Scale. In this study, we noted the clinical significance of endoscope-assisted clipping in preventing PCI. These findings could reduce the incidence of PCI and improve the understanding of its mechanisms of action. However, a larger and longer-term study is required to evaluate the benefits of endoscopy on clinical outcomes.

Fast and accurate interpretation of workload classification model.

How can we interpret predictions of a workload classification model? A workload is a sequence of operations executed in DRAM, where each operation contains a command and an address. Classifying a given sequence into a correct workload type is important for verifying the quality of DRAM. Although a previous model achieves a reasonable accuracy on workload classification, it is challenging to interpret the prediction results since it is a black box model. A promising direction is to exploit interpretation models which compute the amount of attribution each feature gives to the prediction. However, none of the existing interpretable models are tailored for workload classification. The main challenges to be addressed are to 1) provide interpretable features for further improving interpretability, 2) measure the similarity of features for constructing the interpretable super features, and 3) provide consistent interpretations over all instances. In this paper, we propose INFO (INterpretable model For wOrkload classification), a model-agnostic interpretable model which analyzes workload classification results. INFO provides interpretable results while producing accurate predictions. We design super features to enhance interpretability by hierarchically clustering original features used for the classifier. To generate the super features, we define and measure the interpretability-friendly similarity, a variant of Jaccard similarity between original features. Then, INFO globally explains the workload classification model by generalizing super features over all instances. Experiments show that INFO provides intuitive interpretations which are faithful to the original non-interpretable model. INFO also shows up to 2.0× faster running time than the competitor while having comparable accuracies for real-world workload datasets.

Accurate bundle matching and generation via multitask learning with partially shared parameters.

How can we recommend existing bundles to users accurately? How can we generate new tailored bundles for users? Recommending a bundle, or a group of various items, has attracted widespread attention in e-commerce owing to the increased satisfaction of both users and providers. Bundle matching and bundle generation are two representative tasks in bundle recommendation. The bundle matching task is to correctly match existing bundles to users while the bundle generation is to generate new bundles that users would prefer. Although many recent works have developed bundle recommendation models, they fail to achieve high accuracy since they do not handle heterogeneous data effectively and do not learn a method for customized bundle generation. In this paper, we propose BundleMage, an accurate approach for bundle matching and generation. BundleMage effectively mixes user preferences of items and bundles using an adaptive gate technique to achieve high accuracy for the bundle matching. BundleMage also generates a personalized bundle by learning a generation module that exploits a user preference and the characteristic of a given incomplete bundle to be completed. BundleMage further improves its performance using multi-task learning with partially shared parameters. Through extensive experiments, we show that BundleMage achieves up to 6.6% higher nDCG in bundle matching and 6.3× higher nDCG in bundle generation than the best competitors. We also provide qualitative analysis that BundleMage effectively generates bundles considering both the tastes of users and the characteristics of target bundles.

SensiMix: Sensitivity-Aware 8-bit index & 1-bit value mixed precision quantization for BERT compression.

Given a pre-trained BERT, how can we compress it to a fast and lightweight one while maintaining its accuracy? Pre-training language model, such as BERT, is effective for improving the performance of natural language processing (NLP) tasks. However, heavy models like BERT have problems of large memory cost and long inference time. In this paper, we propose SensiMix (Sensitivity-Aware Mixed Precision Quantization), a novel quantization-based BERT compression method that considers the sensitivity of different modules of BERT. SensiMix effectively applies 8-bit index quantization and 1-bit value quantization to the sensitive and insensitive parts of BERT, maximizing the compression rate while minimizing the accuracy drop. We also propose three novel 1-bit training methods to minimize the accuracy drop: Absolute Binary Weight Regularization, Prioritized Training, and Inverse Layer-wise Fine-tuning. Moreover, for fast inference, we apply FP16 general matrix multiplication (GEMM) and XNOR-Count GEMM for 8-bit and 1-bit quantization parts of the model, respectively. Experiments on four GLUE downstream tasks show that SensiMix compresses the original BERT model to an equally effective but lightweight one, reducing the model size by a factor of 8× and shrinking the inference time by around 80% without noticeable accuracy drop.

DAO-CP: Data-Adaptive Online CP decomposition for tensor stream.

How can we accurately and efficiently decompose a tensor stream? Tensor decomposition is a crucial task in a wide range of applications and plays a significant role in latent feature extraction and estimation of unobserved entries of data. The problem of efficiently decomposing tensor streams has been of great interest because many real-world data dynamically change over time. However, existing methods for dynamic tensor decomposition sacrifice the accuracy too much, which limits their usages in practice. Moreover, the accuracy loss becomes even more serious when the tensor stream has an inconsistent temporal pattern since the current methods cannot adapt quickly to a sudden change in data. In this paper, we propose DAO-CP, an accurate and efficient online CP decomposition method which adapts to data changes. DAO-CP tracks local error norms of the tensor streams, detecting a change point of the error norms. It then chooses the best strategy depending on the degree of changes to balance the trade-off between speed and accuracy. Specifically, DAO-CP decides whether to (1) reuse the previous factor matrices for the fast running time or (2) discard them and restart the decomposition to increase the accuracy. Experimental results show that DAO-CP achieves the state-of-the-art accuracy without noticeable loss of speed compared to existing methods.

Signed random walk diffusion for effective representation learning in signed graphs.

How can we model node representations to accurately infer the signs of missing edges in a signed social graph? Signed social graphs have attracted considerable attention to model trust relationships between people. Various representation learning methods such as network embedding and graph convolutional network (GCN) have been proposed to analyze signed graphs. However, existing network embedding models are not end-to-end for a specific task, and GCN-based models exhibit a performance degradation issue when their depth increases. In this paper, we propose Signed Diffusion Network (SidNet), a novel graph neural network that achieves end-to-end node representation learning for link sign prediction in signed social graphs. We propose a new random walk based feature aggregation, which is specially designed for signed graphs, so that SidNet effectively diffuses hidden node features and uses more information from neighboring nodes. Through extensive experiments, we show that SidNet significantly outperforms state-of-the-art models in terms of link sign prediction accuracy.

Pea-KD: Parameter-efficient and accurate Knowledge Distillation on BERT.

Knowledge Distillation (KD) is one of the widely known methods for model compression. In essence, KD trains a smaller student model based on a larger teacher model and tries to retain the teacher model's level of performance as much as possible. However, existing KD methods suffer from the following limitations. First, since the student model is smaller in absolute size, it inherently lacks model capacity. Second, the absence of an initial guide for the student model makes it difficult for the student to imitate the teacher model to its fullest. Conventional KD methods yield low performance due to these limitations. In this paper, we propose Pea-KD (Parameter-efficient and accurate Knowledge Distillation), a novel approach to KD. Pea-KD consists of two main parts: Shuffled Parameter Sharing (SPS) and Pretraining with Teacher's Predictions (PTP). Using this combination, we are capable of alleviating the KD's limitations. SPS is a new parameter sharing method that increases the student model capacity. PTP is a KD-specialized initialization method, which can act as a good initial guide for the student. When combined, this method yields a significant increase in student model's performance. Experiments conducted on BERT with different datasets and tasks show that the proposed approach improves the student model's performance by 4.4% on average in four GLUE tasks, outperforming existing KD baselines by significant margins.

Multi-EPL: Accurate multi-source domain adaptation.

Given multiple source datasets with labels, how can we train a target model with no labeled data? Multi-source domain adaptation (MSDA) aims to train a model using multiple source datasets different from a target dataset in the absence of target data labels. MSDA is a crucial problem applicable to many practical cases where labels for the target data are unavailable due to privacy issues. Existing MSDA frameworks are limited since they align data without considering labels of the features of each domain. They also do not fully utilize the target data without labels and rely on limited feature extraction with a single extractor. In this paper, we propose Multi-EPL, a novel method for MSDA. Multi-EPL exploits label-wise moment matching to align the conditional distributions of the features for the labels, uses pseudolabels for the unavailable target labels, and introduces an ensemble of multiple feature extractors for accurate domain adaptation. Extensive experiments show that Multi-EPL provides the state-of-the-art performance for MSDA tasks in both image domains and text domains, improving the accuracy by up to 13.20%.

Compressing deep graph convolution network with multi-staged knowledge distillation.

Given a trained deep graph convolution network (GCN), how can we effectively compress it into a compact network without significant loss of accuracy? Compressing a trained deep GCN into a compact GCN is of great importance for implementing the model to environments such as mobile or embedded systems, which have limited computing resources. However, previous works for compressing deep GCNs do not consider the multi-hop aggregation of the deep GCNs, though it is the main purpose for their multiple GCN layers. In this work, we propose MustaD (Multi-staged knowledge Distillation), a novel approach for compressing deep GCNs to single-layered GCNs through multi-staged knowledge distillation (KD). MustaD distills the knowledge of 1) the aggregation from multiple GCN layers as well as 2) task prediction while preserving the multi-hop feature aggregation of deep GCNs by a single effective layer. Extensive experiments on four real-world datasets show that MustaD provides the state-of-the-art performance compared to other KD based methods. Specifically, MustaD presents up to 4.21%p improvement of accuracy compared to the second-best KD models.

Unsupervised multi-source domain adaptation with no observable source data.

Given trained models from multiple source domains, how can we predict the labels of unlabeled data in a target domain? Unsupervised multi-source domain adaptation (UMDA) aims for predicting the labels of unlabeled target data by transferring the knowledge of multiple source domains. UMDA is a crucial problem in many real-world scenarios where no labeled target data are available. Previous approaches in UMDA assume that data are observable over all domains. However, source data are not easily accessible due to privacy or confidentiality issues in a lot of practical scenarios, although classifiers learned in source domains are readily available. In this work, we target data-free UMDA where source data are not observable at all, a novel problem that has not been studied before despite being very realistic and crucial. To solve data-free UMDA, we propose DEMS (Data-free Exploitation of Multiple Sources), a novel architecture that adapts target data to source domains without exploiting any source data, and estimates the target labels by exploiting pre-trained source classifiers. Extensive experiments for data-free UMDA on real-world datasets show that DEMS provides the state-of-the-art accuracy which is up to 27.5% point higher than that of the best baseline.

AUBER: Automated BERT regularization.

How can we effectively regularize BERT? Although BERT proves its effectiveness in various NLP tasks, it often overfits when there are only a small number of training instances. A promising direction to regularize BERT is based on pruning its attention heads with a proxy score for head importance. However, these methods are usually suboptimal since they resort to arbitrarily determined numbers of attention heads to be pruned and do not directly aim for the performance enhancement. In order to overcome such a limitation, we propose AUBER, an automated BERT regularization method, that leverages reinforcement learning to automatically prune the proper attention heads from BERT. We also minimize the model complexity and the action search space by proposing a low-dimensional state representation and dually-greedy approach for training. Experimental results show that AUBER outperforms existing pruning methods by achieving up to 9.58% better performance. In addition, the ablation study demonstrates the effectiveness of design choices for AUBER.

Clinical Significance of Acute Kidney Injury in Lung Cancer Patients.

PURPOSE: Acute kidney injury (AKI) in cancer patients is associated with increased morbidity and mortality. The incidence of AKI in lung cancer seems to be relatively higher compared with other solid organ malignancies, although its impact on patient outcomes remains unclear. MATERIALS AND METHODS: The patients newly diagnosed with lung cancer from 2004 to 2013 were enrolled in this retrospective cohort study. The patients were categorized according to the presence and severity of AKI. We compared all-cause mortality and long-term renal outcome according to AKI stage. RESULTS: A total of 3,202 patients were included in the final analysis. AKI occurred in 1,783 (55.7%) patients during the follow-up period, with the majority having mild AKI stage 1 (75.8%). During the follow-up of 2.6±2.2 years, total 1,251 patients (53.7%) were died and 5-year survival rate was 46.9%. We found that both AKI development and severity were independent risk factors for all-cause mortality in lung cancer patients, even after adjustment for lung cancer-specific variables including the stage or pathological type. In addition, patients suffered from more severe AKI tend to encounter de novo chronic kidney disease development, worsening kidney function, and end-stage kidney disease progression. CONCLUSION: In this study, more than half of the lung cancer patients experienced AKI during their diagnosis and treatment period. Moreover, AKI occurrence and more advanced AKI were associated with a higher mortality risk and adverse kidney outcomes.

PACC: Large scale connected component computation on Hadoop and Spark.

A connected component in a graph is a set of nodes linked to each other by paths. The problem of finding connected components has been applied to diverse graph analysis tasks such as graph partitioning, graph compression, and pattern recognition. Several distributed algorithms have been proposed to find connected components in enormous graphs. Ironically, the distributed algorithms do not scale enough due to unnecessary data IO & processing, massive intermediate data, numerous rounds of computations, and load balancing issues. In this paper, we propose a fast and scalable distributed algorithm PACC (Partition-Aware Connected Components) for connected component computation based on three key techniques: two-step processing of partitioning & computation, edge filtering, and sketching. PACC considerably shrinks the size of intermediate data, the size of input graph, and the number of rounds without suffering from load balancing issues. PACC performs 2.9 to 10.7 times faster on real-world graphs compared to the state-of-the-art MapReduce and Spark algorithms.

FlexGraph: Flexible partitioning and storage for scalable graph mining.

How can we analyze large graphs such as the Web, and social networks with hundreds of billions of vertices and edges? Although many graph mining systems have been proposed to perform various graph mining algorithms on such large graphs, they have difficulties in processing Web-scale graphs due to massive communication and I/O costs caused by communication between workers, and reading subgraphs repeatedly. In this paper, we propose FlexGraph, a scalable distributed graph mining method reducing the costs by exploiting properties of real-world graphs. FlexGraph significantly decreases the communication cost, which is the main bottleneck of distributed systems, by exploiting different edge placement policies based on types of vertices. Furthermore, we propose a flexible storage format to reduce I/O costs when reading input graph repeatedly. Experiments show that FlexGraph succeeds in processing up to 64× larger graphs than existing distributed memory-based graph mining methods, and consistently outperforms previous disk-based graph mining methods.

PS-MCL: parallel shotgun coarsened Markov clustering of protein interaction networks.

BACKGROUND: How can we obtain fast and high-quality clusters in genome scale bio-networks? Graph clustering is a powerful tool applied on bio-networks to solve various biological problems such as protein complexes detection, disease module detection, and gene function prediction. Especially, MCL (Markov Clustering) has been spotlighted due to its superior performance on bio-networks. MCL, however, is skewed towards finding a large number of very small clusters (size 1-3) and fails to detect many larger clusters (size 10+). To resolve this fragmentation problem, MLR-MCL (Multi-level Regularized MCL) has been developed. MLR-MCL still suffers from the fragmentation and, in cases, unrealistically large clusters are generated. RESULTS: In this paper, we propose PS-MCL (Parallel Shotgun Coarsened MCL), a parallel graph clustering method outperforming MLR-MCL in terms of running time and cluster quality. PS-MCL adopts an efficient coarsening scheme, called SC (Shotgun Coarsening), to improve graph coarsening in MLR-MCL. SC allows merging multiple nodes at a time, which leads to improvement in quality, time and space usage. Also, PS-MCL parallelizes main operations used in MLR-MCL which includes matrix multiplication. CONCLUSIONS: Experiments show that PS-MCL dramatically alleviates the fragmentation problem, and outperforms MLR-MCL in quality and running time. We also show that the running time of PS-MCL is effectively reduced with parallelization.

S3CMTF: Fast, accurate, and scalable method for incomplete coupled matrix-tensor factorization.

How can we extract hidden relations from a tensor and a matrix data simultaneously in a fast, accurate, and scalable way? Coupled matrix-tensor factorization (CMTF) is an important tool for this purpose. Designing an accurate and efficient CMTF method has become more crucial as the size and dimension of real-world data are growing explosively. However, existing methods for CMTF suffer from lack of accuracy, slow running time, and limited scalability. In this paper, we propose S3CMTF, a fast, accurate, and scalable CMTF method. In contrast to previous methods which do not handle large sparse tensors and are not parallelizable, S3CMTF provides parallel sparse CMTF by carefully deriving gradient update rules. S3CMTF asynchronously updates partial gradients without expensive locking. We show that our method is guaranteed to converge to a quality solution theoretically and empirically. S3CMTF further boosts the performance by carefully storing intermediate computation and reusing them. We theoretically and empirically show that S3CMTF is the fastest, outperforming existing methods. Experimental results show that S3CMTF is up to 930× faster than existing methods while providing the best accuracy. S3CMTF shows linear scalability on the number of data entries and the number of cores. In addition, we apply S3CMTF to Yelp rating tensor data coupled with 3 additional matrices to discover interesting patterns.

Acute kidney injury predicts all-cause mortality in patients with cancer.

BACKGROUND: Acute kidney injury (AKI) is a critical issue in cancer patients because it is not only a morbid complication but also able to interrupt timely diagnostic evaluation or planned optimal treatment. However, the impact of AKI on overall mortality in cancer patients remains unclear. METHODS: We conducted a retrospective cohort study of 67 986 cancer patients, from 2004 to 2013 to evaluate the relationship between AKI and all-cause mortality. We used KDIGO AKI definition and grading system. RESULTS: During 3.9 ± 3.1 years of follow-up, 33.8% of the patients experienced AKI at least once. Among AKI events, stage 1, 2, and 3 was 71.0%, 13.8%, and 15.1%, respectively. AKI incidence was highest in hematologic malignancies, followed by urinary tract cancer, and hepatocellular carcinoma. Male sex, older age, underlying diabetes and hypertension, lower serum albumin and plasma hemoglobin, more frequent radio-contrast exposure, entrance of clinical trials, and receiving chemotherapy were associated with AKI occurrence. AKI development was an independent risk factor for elevated mortality in cancer patients with dose-responsive manner (Stage 1, hazard ratio [HR] 1.183, 95% confidence interval [CI] 1.145-1.221, P < 0.001; Stage 2, HR 1.710, 95% CI 1.629-1.796; Stage 3, HR 2.000, 95% CI 1.910-2.095; No AKI, reference group) even after adjustment. This tendency was reproduced in various cancer types except thyroid cancer and in various treatment modalities, however, not shown in patients with baseline renal dysfunction. CONCLUSION: AKI was an independent risk factor for all-cause mortality in overall cancer patients with dose-responsive manner.

Supervised and extended restart in random walks for ranking and link prediction in networks.

Given a real-world graph, how can we measure relevance scores for ranking and link prediction? Random walk with restart (RWR) provides an excellent measure for this and has been applied to various applications such as friend recommendation, community detection, anomaly detection, etc. However, RWR suffers from two problems: 1) using the same restart probability for all the nodes limits the expressiveness of random walk, and 2) the restart probability needs to be manually chosen for each application without theoretical justification. We have two main contributions in this paper. First, we propose Random Walk with Extended Restart (RWER), a random walk based measure which improves the expressiveness of random walks by using a distinct restart probability for each node. The improved expressiveness leads to superior accuracy for ranking and link prediction. Second, we propose SuRe (Supervised Restart for RWER), an algorithm for learning the restart probabilities of RWER from a given graph. SuRe eliminates the need to heuristically and manually select the restart parameter for RWER. Extensive experiments show that our proposed method provides the best performance for ranking and link prediction tasks.