Adaptive Hypergraph Auto-Encoder for Relational Data Clustering

Hu, Youpeng; Li, Xunkai; Wang, Yujie; Wu, Yixuan; Zhao, Yining; Yan, Chenggang; Yin, Jian; Gao, Yue

doi:10.1109/tkde.2021.3108192

Cited by 19 publications

(4 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This unwanted case would cause the effect that the embeddings alone do not contain any useful information about the LFP shape, because it would be possible to train further auto-encoder with arbitrary permutations of the embeddings, which perform equally good. However, in order to allow an interpretation of clusters in embedding space, neighbourhood relations between different LFP shapes should be conserved in the embedding layer [71]. Thus, in our study we used shallow-networks to reduce the number of parameters, and added drop-out layers to prevent the neural network from overfitting.…”

Section: Discussionmentioning

confidence: 99%

Deep learning based decoding of local field potential events

Schilling

Gerum

Boehm

et al. 2022

Preprint

View full text Add to dashboard Cite

How is information processed in the cerebral cortex? To answer this question a lot of effort has been undertaken to create novel and to further develop existing neuroimaging techniques. Thus, a high spatial resolution of fMRI devices was the key to exactly localize cognitive processes. Furthermore, an increase in time-resolution and number of recording channels of electro-physiological setups has opened the door to investigate the exact timing of neural activity. However, in most cases the recorded signal is averaged over many (stimulus) repetitions, which erases the fine-structure of the neural signal. Here, we show that an unsupervised machine learning approach can be used to extract meaningful information from electro-physiological recordings on a single-trial base. We use an auto-encoder network to reduce the dimensions of single local field potential (LFP) events to create interpretable clusters of different neural activity patterns. Strikingly, certain LFP shapes correspond to latency differences in different recording channels. Hence, LFP shapes can be used to determine the direction of information flux in the cerebral cortex. Furthermore, after clustering, we decoded the cluster centroids to reverse-engineer the underlying prototypical LFP event shapes. To evaluate our approach, we applied it to both neural extra-cellular recordings in rodents, and intra-cranial EEG recordings in humans. Finally, we find that single channel LFP event shapes during spontaneous activity sample from the realm of possible stimulus evoked event shapes. A finding which so far has only been demonstrated for multi-channel population coding.

show abstract

Section: Discussionmentioning

confidence: 99%

Deep learning based decoding of local field potential events

Schilling

Gerum

Boehm

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…HGNN + [23] introduces higher-order multimodal modeling based on HGNN to learn the optimal solution in a single hypergraph and fuse the correlations of different modalities into a unified hypergraph. AHGAE [24] combines adaptive hypergraph Laplacian smoothing filtering and relational reconstruction self-encoder to ensure that the potential representation can retain both original and higher-order information. DeepHGSL [25] proposes the hypergraph information bottleneck principle (HIB), which is used to construct loss functions to enhance correct connections and weaken incorrect connections during training.…”

Section: Hypergraph Convolutionmentioning

confidence: 99%

Hypergraph network embedding for community detection

Xiang,

You,

Wang

et al. 2024

J Supercomput

View full text Add to dashboard Cite

Using attribute graphs for node embedding to detect community structure has become a popular research topic. However, most of the existing algorithms mainly focus on the network structure and node features, which ignore the higher-order relationships between nodes. In addition, only adopting the original graph structure will suffer from sparsity problems, and will also result in sub-optimal node clustering performance. In this paper, we propose a hypergraph network embedding (HGNE) for community detection to solve the above problems. Firstly, we construct potential connections based on the shared feature information of the nodes. By fusing the original topology with feature-based potential connections, both the explicit and implicit relationships are encoded into the node representations, thus alleviating the sparsity problem. Secondly, for integrating the higher-order relationship, we adopt hypergraph convolution to encode the higherorder correlations. To constrain the quality of the node embedding, the spectral hypergraph embedding loss is utilized. Furthermore, we design a dual-contrast mechanism, which draws similar nodes closer by comparing the representations of different views. This mechanism can efficiently prevent multi-node classes from distorting less-node classes. Finally, the dual-contrast mechanism is jointly optimized with self-training clustering to obtain more robust node representations, thus improving the clustering results. Extensive experiments on five datasets indicate the superiority and effectiveness of HGNE.

show abstract

“…In recent years, Graph Neural Networks (GNNs) [44], [45] and their variants [46], [47], [48], [49] have emerged as powerful tools in the realm of data analysis, demonstrating their versatility in a broad spectrum of graph-structured tasks, including graph classification [50], [51], [52], graph clustering [53], [54], [55], and graph link prediction [56], [57]. The power of GNNs also extends beyond graph-structured data, as they have also been effectively utilized in nongraph structured data.…”

Section: Graph and Hypergraph Neural Networkmentioning

confidence: 99%

Hypergraph Neural Network for Skeleton-Based Action Recognition

Hao

Guo

et al. 2021

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

Action recognition from video data forms a cornerstone with wide-ranging applications. Single-view action recognition faces limitations due to its reliance on a single viewpoint. In contrast, multi-view approaches capture complementary information from various viewpoints for improved accuracy. Recently, event cameras have emerged as innovative bio-inspired sensors, leading to advancements in event-based action recognition. However, existing works predominantly focus on single-view scenarios, leaving a gap in multi-view event data exploitation, particularly in challenges like information deficit and semantic misalignment. To bridge this gap, we introduce HyperMV, a multi-view event-based action recognition framework. HyperMV converts discrete event data into frame-like representations and extracts view-related features using a shared convolutional network. By treating segments as vertices and constructing hyperedges using rule-based and KNN-based strategies, a multi-view hypergraph neural network that captures relationships across viewpoint and temporal features is established. The vertex attention hypergraph propagation is also introduced for enhanced feature fusion. To prompt research in this area, we present the largest multi-view event-based action dataset THU MV-EACT -50, comprising 50 actions from 6 viewpoints, which surpasses existing datasets by over tenfold. Experimental results show that HyperMV significantly outperforms baselines in both cross-subject and cross-view scenarios, and also exceeds the state-of-the-arts in frame-based multi-view action recognition.

show abstract

Adaptive Hypergraph Auto-Encoder for Relational Data Clustering

Cited by 19 publications

References 28 publications

Deep learning based decoding of local field potential events

Deep learning based decoding of local field potential events

Hypergraph network embedding for community detection

Hypergraph Neural Network for Skeleton-Based Action Recognition

Contact Info

Product

Resources

About