Feature selection is a key step in machine learning by eliminating features that are not related to the modeling target to create reliable and interpretable models. By exploring the potential complex correlations among features of unlabeled data, recently introduced self-supervision-enhanced feature selection greatly reduces the reliance on the labeled samples. However, they are generally based on the autoencoder with sample-wise self-supervision, which can hardly exploit the relations among samples. To address this limitation, this article proposes graph representation learning enhanced semi-supervised feature selection (G-FS) which performs feature selection based on the discovery and exploitation of the non-Euclidean relations among features and samples by translating unlabeled “plain” tabular data into a bipartite graph. A self-supervised edge prediction task is designed to distill rich information on the graph into low-dimensional embeddings, which remove redundant features and noise. Guided by the condensed graph representation, we propose a batch attention feature weight generation mechanism that generates more robust weights according to batch-based selection patterns rather than individual samples. The results show that G-FS achieves significant performance edges in 14 datasets compared to twelve state-of-the-art baselines, including two recent self-supervised baselines. The source code is public available at https://github.com/Icannotnamemyselff/G-FS_Graph_enhacned_feature_selection .

Unsupervised graph representation learning aims to condense graph information into dense vector embeddings to support various downstream tasks. To achieve this goal, existing UGRL approaches mainly adopt the message-passing mechanism to simultaneously incorporate graph topology and node attribute with an aggregated view. However, recent research points out that this direct aggregation may lead to issues such as over-smoothing and/or topology distortion, as topology and node attribute of totally different semantics. To address this issue, this paper proposes a novel Graph Dual-view AutoEncoder framework (GDAE) which introduces the node-wise view for an individual node beyond the traditional aggregated view for aggregation of connected nodes. Specifically, the node-wise view captures the unique characteristics of individual node through a decoupling design, i.e., topology encoding by multi-steps random walk while preserving node-wise individual attribute. Meanwhile, the aggregated view aims to better capture the collective commonality among long-range nodes through an enhanced strategy, i.e., topology masking then attribute aggregation. Extensive experiments on 5 synthetic and 11 real-world benchmark datasets demonstrate that GDAE achieves the best results with up to 49.5% and 21.4% relative improvement in node degree prediction and cut-vertex detection tasks and remains top in node classification and link prediction tasks.

Unsupervised graph representation learning (GRL) aims at distilling diverse graph information into task-agnostic embeddings without label supervision. Due to a lack of support from labels, recent representation learning methods usually adopt self-supervised learning, and embeddings are learned by solving a handcrafted auxiliary task (so-called pretext task). However, partially due to the irregular non-Euclidean data in graphs, the pretext tasks are generally designed under homophily assumptions and cornered in the low-frequency signals, which results in significant loss of other signals, especially high-frequency signals widespread in graphs with heterophily. Motivated by this limitation, we propose a multi-view perspective and the usage of diverse pretext tasks to capture different signals in graphs into embeddings. A novel framework, denoted as Multi-view Graph Encoder (MVGE), is proposed, and a set of key designs are identified. More specifically, a set of new pretext tasks are designed to encode different types of signals, and a straightforward operation is proposed to maintain both the commodity and personalization in both the attribute and the structural levels. Extensive experiments on synthetic and real-world network datasets show that the node representations learned with MVGE achieve significant performance improvements in three different downstream tasks, especially on graphs with heterophily.

Unsupervised graph representation learning aims to distill various graph information into a downstream task-agnostic dense vector embedding. However, existing graph representation learning approaches are largely designed under the node homophily assumption: connected nodes tend to have similar labels and aim to optimize performance on node-centric downstream tasks. Their design apparently against the task-agnostic principle and generally suffer poor performance in tasks, e.g., edge classification task, that demands feature signals beyond both the node-view and homophily assumption. To condense different feature signals into the edge embeddings, this paper proposes PairE, a novel unsupervised graph embedding method using two paired nodes as the basic unit of embedding to retain the high-frequency signals between nodes to support both node-related and edge-related tasks. Accordingly, a multi-self-supervised autoencoder is designed to fulfill two pretext tasks: one retains the high-frequency signal better, and another enhances the representation of commonality. Our extensive experiments on a diversity of benchmark datasets clearly show that PairE outperforms the unsupervised state-of-the-art baselines, with up to 81% improvement on the edge classification tasks that rely on both the high and low-frequency signals in the pair and up to 42% performance gain on the node classification tasks.

Reducing sensor requirements while keeping optimal control performance is crucial to many industrial control applications to achieve robust, low-cost, and computation-efficient controllers. However, existing feature selection solutions for the typical machine learning domain can hardly be applied in the domain of control with changing dynamics. In this paper, a novel framework, namely the Dual-world embedded Attentive Feature Selection (D-AFS), can efficiently select the most relevant sensors for the system under dynamic control. Rather than the one world used in most Deep Reinforcement Learning (DRL) algorithms, D-AFS has both the real world and its virtual peer with twisted features. By analyzing the DRL's response in two worlds, D-AFS can quantitatively identify respective features' importance towards control. A well-known active flow control problem, cylinder drag reduction, is used for evaluation. Results show that D-AFS successfully finds an optimized five-probes layout with 18.7% drag reduction than the state-of-the-art solution with 151 probes and 49.2% reduction than five-probes layout by human experts. We also apply this solution to four OpenAI classical control cases. In all cases, D-AFS achieves the same or better sensor configurations than originally provided solutions. Results highlight, we argued, a new way to achieve efficient and optimal sensor designs for experimental or industrial systems. Our source codes are made publicly available at https://github.com/G-AILab/DAFSFluid.