Author name cluster

Jun Yuan

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

12 papers

2 author rows

EAAI Journal 2025 Journal Article

Hierarchical cascaded networks with multi-task balanced loss for fine-grained hashing

Shun Liu
Yanjun Zheng
Xianxian Zeng
Jun Yuan
Jiawen Li
Rongjun Chen

Fine-grained image retrieval has seen significant advancements with the rise of deep hashing methods. However, these methods often prioritize high-level features, which may lead to the loss of important low-level details in hash code representations. Additionally, balancing the classification and hashing tasks remains a challenge. To address these issues, we propose a Hierarchical Cascaded Network (HCN) with a multi-task balanced loss function tailored for fine-grained hashing. Our model captures detailed information from different feature levels through a hierarchical backbone network and utilizes a cascaded representation learning module to enhance and fuse features using attention mechanisms. An adaptive loss function ensures a balanced contribution from both classification and hashing tasks during training. Extensive experiments on benchmark datasets demonstrate that HCN outperforms state-of-the-art methods, achieving a promising improvement across five datasets and multiple hash code lengths. These results highlight the effectiveness of HCN in enhancing fine-grained image retrieval, with potential applications in areas requiring both high accuracy and efficient retrieval.

Details DOI

TCS Journal 2025 Journal Article

The local diagnosability of directed interconnection networks

Aixia Liu
Shuchang Chai
Lina Li
Chenhui Liang
Jun Yuan

The diagnosability is a vital parameter to measure the self-fault diagnosis ability of interconnection networks of computer systems. The local diagnosability is a generalization of diagnosability, which focuses on the fault diagnosability at a given processor. In this paper, we discuss the local diagnosability and the good in-neighbor conditional local diagnosability of directed networks, and present some necessary and sufficient conditions for a directed network to be locally ζ-diagnosable or good in-neighbor conditional locally ζ-diagnosable at a processor. We also present an algorithm under the PMC model to determine the fault or fault-free state of a given processor in a directed network. As an empirical testing, we demonstrate that our algorithm can ascertain the state of each vertex of unidirectional hypercube U Q n if the number of faulty vertices is not more than n − 1.

Details DOI

AAAI Conference 2024 Conference Paper

MESED: A Multi-Modal Entity Set Expansion Dataset with Fine-Grained Semantic Classes and Hard Negative Entities

Yangning Li
Tingwei Lu
Hai-Tao Zheng
Yinghui Li
Shulin Huang
Tianyu Yu
Jun Yuan
Rui Zhang

The Entity Set Expansion (ESE) task aims to expand a handful of seed entities with new entities belonging to the same semantic class. Conventional ESE methods are based on mono-modality (i.e., literal modality), which struggle to deal with complex entities in the real world such as (1) Negative entities with fine-grained semantic differences. (2) Synonymous entities. (3) Polysemous entities. (4) Long-tailed entities. These challenges prompt us to propose novel Multi-modal Entity Set Expansion (MESE), where models integrate information from multiple modalities to represent entities. Intuitively, the benefits of multi-modal information for ESE are threefold: (1) Different modalities can provide complementary information. (2) Multi-modal information provides a unified signal via common visual properties for the same semantic class or entity. (3) Multi-modal information offers robust alignment signals for synonymous entities. To assess model performance in MESE, we constructed the MESED dataset which is the first multi-modal dataset for ESE with large-scale and elaborate manual calibration. A powerful multi-modal model MultiExpan is proposed which is pre-trained on four multimodal pre-training tasks. The extensive experiments and analyses on MESED demonstrate the high quality of the dataset and the effectiveness of our MultiExpan, as well as pointing the direction for future research. The benchmark and code are public at https://github.com/THUKElab/MESED.

PDF Details DOI

IJCAI Conference 2023 Conference Paper

JEPOO: Highly Accurate Joint Estimation of Pitch, Onset and Offset for Music Information Retrieval

Haojie Wei
Jun Yuan
Rui Zhang
Yueguo Chen
Gang Wang

Melody extraction is a core task in music information retrieval, and the estimation of pitch, onset and offset are key sub-tasks in melody extraction. Existing methods have limited accuracy, and work for only one type of data, either single-pitch or multi-pitch. In this paper, we propose a highly accurate method for joint estimation of pitch, onset and offset, named JEPOO. We address the challenges of joint learning optimization and handling both single-pitch and multi-pitch data through novel model design and a new optimization technique named Pareto modulated loss with loss weight regularization. This is the first method that can accurately handle both single-pitch and multi-pitch music data, and even a mix of them. A comprehensive experimental study on a wide range of real datasets shows that JEPOO outperforms state-of-the-art methods by up to 10. 6\%, 8. 3\% and 10. 3\% for the prediction of Pitch, Onset and Offset, respectively, and JEPOO is robust for various types of data and instruments. The ablation study validates the effectiveness of each component of JEPOO.

PDF Details DOI

TCS Journal 2023 Journal Article

The partial diagnosability of interconnection networks under the Hybrid PMC model

Jun Yuan
Shuyuan Ge
Aixia Liu

System level diagnosis is a primary method to identify the faulty elements in a multiprocessor system. Different from the traditional system level fault diagnosis models, the new fault diagnosis model, hybrid PMC model (short for HPMC model), is suitable to the circumstances that the processor and link faults occur simultaneously. Under the HPMC model, the q-edge restricted diagnosability of a system H, denoted by t q e ( H ), is the maximum number p such that H can correctly identify all the faulty elements containing p faulty processors and at most q faulty links. Similarly, the p-vertex restricted edge-diagnosability of a system H, denoted by s p v ( H ), is the maximum number q such that H can correctly identify all the faulty elements containing q faulty links and at most p faulty processors. These two diagnosability are collectively referred to as partial diagnosability. We investigate and determine the q-edge restricted diagnosability of a system H under the HPMC model for q ≥ η ( H ), where η ( H ) = max ⁡ { | N ( u ) ∩ N ( v ) | | u v ∈ E ( H ) }. In addition, we investigate the relationship between t q e ( H ) and s p v ( H ) under the HPMC model, and show that under some conditions, s p v ( H ) = q if t q e ( H ) = p. Applying our results, the two kinds of partial diagnosability of k-ary n-cubes, bijective connection networks and undirected Kautz graphs are established.

Details DOI

TCS Journal 2022 Journal Article

The non-inclusive g-good-neighbor diagnosability of interconnection networks

Jun Yuan
Ying Li
Aixia Liu
Huijuan Qiao

Motivated by the definitions of g-good-neighbor diagnosability and non-inclusive diagnosability, we propose a new diagnosability—the non-inclusive g-good-neighbor diagnosability t N g ( G ) of a multiprocessor system G, which requires every pair of g-good-neighbor faulty sets is non-inclusive. The R g -conditional diagnosability t R g ( G ) of a system G is a generalization of conditional diagnosability, which requires at least g fault-free neighbors for each node. In this paper, we explore the relationships between the non-inclusive g-good-neighbor diagnosability and the R g -conditional diagnosability of G under the PMC and MM* models. We first show t N g ( G ) ≤ t R g ( G ) for g ≥ 1, and also give some conditions for equality. Next, we discuss the non-inclusive g-good-neighbor diagnosability of hypercubes, ( n, k ) -star graphs and ( n, k ) -bubble-sort graphs. We show that the non-inclusive g-good-neighbor diagnosability of n-dimensional hypercubes is less that its R g -conditional diagnosability for 2 ≤ g ≤ n − 2 2, and determine the non-inclusive g-good-neighbor diagnosability of ( n, k ) -star graphs and ( n, k ) -bubble-sort graphs. Finally, we plot and compare the non-inclusive g-good-neighbor diagnosability and the g-good-neighbor diagnosability of ( n, k ) -star graphs and ( n, k ) -bubble-sort graphs under the PMC and MM* models, respectively. It can be seen that their non-inclusive g-good-neighbor diagnosability is significantly larger than their g-good-neighbor diagnosability.

Details DOI

TCS Journal 2022 Journal Article

The R-conditional diagnosability of international networks

Jun Yuan
Huijuan Qiao
Aixia Liu

The R g -conditional diagnosability of a multiprocessor system modeled by a graph G, denoted by t R g ( G ), is a generalization of conditional diagnosability, which restricts every vertex contains at least g fault-free neighbors. Particularly, the R 1 -conditional diagnosability is the conditional diagnosability. The R g -conditional connectivity of a graph G, denoted by κ R g ( G ), is the minimum number of vertices, whose deletion will disconnect the graph and every vertex of G has at least g neighbors in the remaining subgraphs. In this paper, the relationships between the R g -conditional connectivity of a graph G and its R g -conditional diagnosability under the PMC and MM* models are explored. We establish the R g -conditional diagnosability t R g ( G ) equals κ R 2 g + 1 ( G ) + g under some reasonable conditions, except the R 1 -conditional diagnosability of G under the MM* model. Moreover, we show under the MM* model, t R 1 ( G ) = κ R 2 ( G ) with similar conditions. Applying our results, the R g -conditional diagnosability of the ( n, k ) -star graphs and the ( n, k ) -bubble-sort graphs are determined.

Details DOI

ICLR Conference 2021 Conference Paper

Mind the Pad - CNNs Can Develop Blind Spots

Bilal Alsallakh
Narine Kokhlikyan
Vivek Miglani
Jun Yuan
Orion Reblitz-Richardson

We show how feature maps in convolutional networks are susceptible to spatial bias. Due to a combination of architectural choices, the activation at certain locations is systematically elevated or weakened. The major source of this bias is the padding mechanism. Depending on several aspects of convolution arithmetic, this mechanism can apply the padding unevenly, leading to asymmetries in the learned weights. We demonstrate how such bias can be detrimental to certain tasks such as small object detection: the activation is suppressed if the stimulus lies in the impacted area, leading to blind spots and misdetection. We explore alternative padding methods and propose solutions for analyzing and mitigating spatial bias.

Details

TCS Journal 2019 Journal Article

The h-extra connectivity of k-ary n-cubes

Aixia Liu
Shiying Wang
Jun Yuan
Xue Ma

Reliability evaluation of interconnection network is important to the design and maintenance of multiprocessor systems. The h-extra connectivity κ h ( G ) is an important subject for a multiprocessor system's reliability to tolerate faulty processors. It is defined as the minimum cardinality of a set of vertices in G, whose deletion disconnects G and leaves every remaining component with more than h vertices. In this paper, we investigate the h-extra connectivity of k-ary n-cube Q n k, a well-known interconnection network proposed for multiprocessor systems, and show κ h ( Q n k ) = { ( h + 1 ) 2 n − 2 h − ( h 2 ), if 0 ≤ h ≤ n, n ≥ 2, k ≥ 4; ( h + 1 ) 2 n − 3 h − ( h 2 ), if 0 ≤ h ≤ n, n ≥ 3, k = 3, except κ 2 ( Q 2 4 ) = 6.

Details DOI

AAAI Conference 2019 Conference Paper

TransGate: Knowledge Graph Embedding with Shared Gate Structure

Jun Yuan
Neng Gao
Ji Xiang

Embedding knowledge graphs (KGs) into continuous vector space is an essential problem in knowledge extraction. Current models continue to improve embedding by focusing on discriminating relation-specific information from entities with increasingly complex feature engineering. We noted that they ignored the inherent relevance between relations and tried to learn unique discriminate parameter set for each relation. Thus, these models potentially suffer from high time complexity and large parameters, preventing them from efficiently applying on real-world KGs. In this paper, we follow the thought of parameter sharing to simultaneously learn more expressive features, reduce parameters and avoid complex feature engineering. Based on gate structure from LSTM, we propose a novel model TransGate and develop shared discriminate mechanism, resulting in almost same space complexity as indiscriminate models. Furthermore, to develop a more effective and scalable model, we reconstruct the gate with weight vectors making our method has comparative time complexity against indiscriminate model. We conduct extensive experiments on link prediction and triplets classification. Experiments show that TransGate not only outperforms state-of-art baselines, but also reduces parameters greatly. For example, TransGate outperforms ConvE and R- GCN with 6x and 17x fewer parameters, respectively. These results indicate that parameter sharing is a superior way to further optimize embedding and TransGate finds a better trade-off between complexity and expressivity.

PDF Details

TCS Journal 2017 Journal Article

On g-extra conditional diagnosability of hypercubes and folded hypercubes

Aixia Liu
Shiying Wang
Jun Yuan
Jing Li

Diagnosability of a multiprocessor system is one important study topic, which plays an important role in measuring of the reliability of multiprocessor systems. In the work of Zhang et al. in 2016, they proposed a new measure for fault diagnosis of systems, namely, g-extra conditional diagnosability. It is defined as the diagnosability of a multiprocessor system under the assumption that every fault-free component contains more than g vertices, which can measure the reliability of interconnection networks in heterogeneous environments more accurately than traditional diagnosability. As two kind of favorable topology structures of interconnection networks, the n-dimensional hypercubes Q n and folded hypercubes F Q n have many good properties. In this paper, we investigate their g-extra conditional diagnosability and show that (a) the g-extra conditional diagnosability of Q n is ( g + 1 ) n − g − C g 2 for n ≥ 5 and 1 ≤ g ≤ n − 1 4 under the MM* model; (b) the g-extra conditional diagnosability of F Q n is ( g + 1 ) n − C g 2 + 1 for n ≥ 9 and 1 ≤ g ≤ n 4 under the MM* model.

Details DOI

TCS Journal 2016 Journal Article

g-Good-neighbor conditional diagnosability measures for 3-ary n-cube networks

Jun Yuan
Aixia Liu
Xiao Qin
Jifu Zhang
Jing Li

The diagnosability of a parallel system is defined as the maximum number of faulty processors or nodes that the system can guarantee to identify. In this study, we investigate the g-good-neighbor conditional diagnosability, which indicates that every fault-free node in a system contains at least g fault-free neighbors. Compared with the conventional diagnosability, g-good-neighbor conditional diagnosability improves accuracy in measuring the reliability of interconnection networks in heterogeneous environments. We apply the PMC and MM* models to study the g-good-neighbor conditional diagnosability of 3-ary n-cube networks, which represent a family of popular parallel systems such as IBM's Blue Gene and Cray T3D. The findings made in this study facilitate accurate reliability measurements in modern parallel systems powered by 3-ary n-cube networks. Specifically, our results show that the g-good-neighbor conditional diagnosability of 3-ary n-cube is g 2 ( 2 n − g + 1 ) − 1 and g − 1 2 ( 4 n − 2 g + 1 ) − 1 when the g value is even and odd, respectively.

Details DOI