Abstract:Graph filters are a staple tool for processing signals over graphs in a multitude of downstream tasks. However, they are commonly designed for graphs with a fixed number of nodes, despite real-world networks typically grow over time. This topological evolution is often known up to a stochastic model, thus, making conventional graph filters ill-equipped to withstand such topological changes, their uncertainty, as well as the dynamic nature of the incoming data. To tackle these issues, we propose an online graph filtering framework by relying on online learning principles. We design filters for scenarios where the topology is both known and unknown, including a learner adaptive to such evolution. We conduct a regret analysis to highlight the role played by the different components such as the online algorithm, the filter order, and the growing graph model. Numerical experiments with synthetic and real data corroborate the proposed approach for graph signal inference tasks and show a competitive performance w.r.t. baselines and state-of-the-art alternatives.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: Most of the existing graph filters are designed for graphs with a fixed number of nodes, but networks in the real world usually grow over time. The evolution of this topological structure often follows some random model, making it difficult for traditional graph filters to deal with these topological changes, uncertainties, and dynamically changing data. To address these issues, the author proposes an online graph filtering framework based on the principle of online learning.
### Specific description of the problem
1. **Stream - based data processing**:
- Data arrives in the form of a stream, that is, new nodes keep joining the graph instead of all node information being provided at once. This requires that the filter can be updated in real - time rather than relying on batch processing.
2. **Dynamic changes in topological structure**:
- The topological structure of the graph may change slowly or rapidly, affecting the design of the online filter. Specifically, how new nodes are connected to the existing graph may be known (deterministic scenario) or unknown (random scenario).
3. **Uncertainty in data distribution**:
- The data on newly added nodes does not necessarily follow a known distribution, so a filter design that is suitable for the current task is required. Sometimes we don't even know how new nodes are connected to the existing graph, for example, in a cold - start recommendation system.
### Solution
The author proposes an online graph filtering framework with the following main contributions:
1. **Online filter design framework**:
- Model the inference problem as a time - varying loss function and update the filter parameters through the online learning principle.
2. **Adapt to two scenarios**:
- **Deterministic scenario**: The connection information of new nodes is known.
- **Random scenario**: The connection information of new nodes is unknown. For these two scenarios, the author conducts regret analysis, discussing the impact of new node connections and the role of graph filters.
3. **Online integration and adaptive random update**:
- In the random scenario, in addition to the filter parameters, the combination parameters of different random connection rules are also learned. The author also discusses the regret analysis in this scenario and analyzes the impact of the integration method on regret.
### Experimental verification
The author verifies the effectiveness of the proposed method through numerical experiments on synthetic data and real - world data (such as recommendation systems and COVID case prediction). The results show that the proposed online filter outperforms other baseline methods and state - of - the - art alternatives in performance, and can perform well especially in random scenarios.
### Formula summary
- **Graph signal representation**:
\[
x_t=[x_{t - 1},x_t]^{\top}
\]
where \(x_t\) is the signal on the new node \(v_t\).
- **Filter output**:
\[
\tilde{y}_t=\sum_{k = 0}^K h_k A_t^k\tilde{x}_t
\]
- **Loss function in the deterministic scenario**:
\[
l_t(h,x_t)=\frac{1}{2}(a_t^{\top}A_{x,t - 1}h - x_t)^2+\mu\|h\|_2^2
\]
- **Loss function in the random scenario**:
\[
l_t(h,x_t)=\mathbb{E}\left[\frac{1}{2}(a_t^{\top}A_{x,t - 1}h - x_t)^2\right]+\mu\|h\|_2^2
\]
Through the above methods, the author successfully solves the application challenges of graph filters on dynamic graphs and improves their robustness and effectiveness in practical application scenarios.