Open-world structured sequence learning via dense target encoding
Structured sequences are popularly used to describe graph data with time-evolving node features and edges. A typical real-world scenario of structured sequences is that unknown class labels continuously arrive and thus the training and testing often across different class spaces. This scenario is al...
Uloženo v:
| Vydáno v: | Information sciences Ročník 680; s. 121147 |
|---|---|
| Hlavní autoři: | , , , , , , , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Elsevier Inc
01.10.2024
|
| Témata: | |
| ISSN: | 0020-0255 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | Structured sequences are popularly used to describe graph data with time-evolving node features and edges. A typical real-world scenario of structured sequences is that unknown class labels continuously arrive and thus the training and testing often across different class spaces. This scenario is also referred to as the open-world learning problem on structured sequences. In this paper, we present a new Dense Open-world Structured Sequence Learning model (DOSSL for short) to learn graph streams in the open-world learning setting. To capture both structural and temporal information, DOSSL uses a GNN-based stochastic recurrent neural network for learning node representation in graph streams, then a truncated Laplacian distribution to describe the latent distribution of graph nodes, and a sampling function is used to generate node representations. Further, DOSSL learns dense target embeddings for the known classes to improve the compactness of known class distribution and reserve enough space for open-world unknown classes. The ultimate open-world classifier is optimized to detect the samples from unknown classes under the constraints of DVAE loss, label loss, class uncertainty loss, and dense target loss. Through empirical analysis conducted on real-world datasets, it has been demonstrated that the advanced technique known as DOSSL exhibits the ability to acquire precise node classifiers by harnessing the power of graph streams.
•We represent an initial endeavor in investigating the open-world learning problem within the context of graph streams. To address this challenge, we introduce a novel dense open-world structured sequence learning model, DOSSL, as a proposed solution.•We effectively solve the technical obstacles pertaining to the temporal and structural dynamics, as well as the fluctuating class labels observed in open-world graph streams. DOSSL employs a recurrent neural network based on GCN, which enables the capturing of both temporal and structural dynamics. To enhance the learning process, stochastic states are introduced alongside the conventional deterministic states within the Gaussian distribution. The stochastic components enable the learning of a latent probabilistic model for each node at every time step.•We enhance the proposed DOSSL model with the learning of dense target embedding. It can change the representation of the target classes and better match the known class space. According to the number of different topological spaces enabled by the type of encoding, dense target encoding avoids the limitation of the space complexity represented by the one-hot target encoding. |
|---|---|
| ISSN: | 0020-0255 |
| DOI: | 10.1016/j.ins.2024.121147 |