Lingfeng Niu’s research while affiliated with Chinese Academy of Sciences and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (96)


A universal network strategy for lightspeed computation of entropy-regularized optimal transport
  • Article

December 2024

·

4 Reads

Neural Networks

·

·

·

[...]

·

Lingfeng Niu





A unified pre-training and adaptation framework for combinatorial optimization on graphs

February 2024

·

11 Reads

Science China Mathematics

Combinatorial optimization (CO) on graphs is a classic topic that has been extensively studied across many scientific and industrial fields. Recently, solving CO problems on graphs through learning methods has attracted great attention. Advanced deep learning methods, e.g., graph neural networks (GNNs), have been used to effectively assist the process of solving COs. However, current frameworks based on GNNs are mainly designed for certain CO problems, thereby failing to consider their transferable and generalizable abilities among different COs on graphs. Moreover, simply using original graphs to model COs only captures the direct correlations among objects, which does not consider the mathematical logicality and properties of COs. In this paper, we propose a unified pre-training and adaptation framework for COs on graphs with the help of the maximum satisfiability (Max-SAT) problem. We first use Max-SAT to bridge different COs on graphs since they can be converted to Max-SAT problems represented by standard formulas and clauses with logical information. Then we further design a pre-training and domain adaptation framework to extract the transferable and generalizable features so that different COs can benefit from them. In the pre-training stage, Max-SAT instances are generated to initialize the parameters of the model. In the fine-tuning stage, instances from CO and Max-SAT problems are used for adaptation so that the transferable ability can be further improved. Numerical experiments on several datasets show that features extracted by our framework exhibit superior transferability and Max-SAT can boost the ability to solve COs on graphs.






Citations (61)


... In many instances, less than 1 percent of the inputs exhibit non-zero values. AI tools consider such raw material as sparse [26], emphasizing the scattered nature of the entire dataset [27]. ...

Reference:

Unlocking Online Insights: LSTM Exploration and Transfer Learning Prospects
Sparse optimization guided pruning for neural networks
  • Citing Article
  • March 2024

Neurocomputing

... The VRP is classified as a non-polynomial hard problem. After the first mention, the problem was investigated for diverse applications, changing through the years with formulations more robust and realistic than previously [15]. The main constraints to solve the problem are related to the vehicle's capacity and time windows to perform the service. ...

A Brief Survey on Learning Based Methods for Vehicle Routing Problems
  • Citing Article
  • January 2023

Procedia Computer Science

... These analysis tools uncover hidden information and provide a better understanding of network dynamics. The graph partitioning problem is a classic combinatorial optimization problem that has been the subject of much research [78]. It aims to divide a graph in two subgraphs by minimizing the total cost of the cut edges. ...

Semi-Supervised Graph Neural Networks for Graph Partitioning Problem
  • Citing Article
  • January 2023

Procedia Computer Science

... The y i is the true label and y i is the predicted probability distribution. The equation 3 is L1 regularization loss, it represents the sum of the absolute values of the model parameters and is used to prevent overfitting [40]. By incorporating the L1 regularization term into the loss function, the model is encouraged to produce sparse parameters, meaning that more parameters are driven toward zero. ...

Federated Learning With l1 Regularization
  • Citing Article
  • May 2023

Pattern Recognition Letters

... However, the latent weights oscillating around the center of quantization bin causes implicit stochasticity during the binary training. As shown in Figure 1, we observe that the learned latent weights tend to cluster around their mean centers, and the distribution exhibits a pronounced peak near 0, which coincides with previous observations [20,44]. We argue that the closer the latent weights are to the point 0 (i.e., the center of the quantized bin), the more easily their gradients are reversed, leading to the weights oscillation 1 Thus, it requires a regularization scheme to induce a bimodal distribution of weights that can solve the oscillation problem. ...

Diluted Binary Neural Network
  • Citing Article
  • March 2023

Pattern Recognition

... Addressing client drift issues involves employing model aggregation and model training methods [10]. FedAvgM [11] introduces momentum on top of Stochastic Gradient Descent (SGD) for model aggregation, a concept similarly utilized in FedMIM [12]. ...

Optimization Strategies for Client Drift in Federated Learning: A review
  • Citing Article
  • December 2022

Procedia Computer Science

... In this context, many deep learning approaches have been proposed since the emergence of the disease. These strategies show promise for automating COVID-19 diagnosis and reducing the strain on medical systems [10]. Several researchers analyzed both chest X-ray and CT scan images using computer vision approaches that presented good results. ...

A lightweight network for COVID-19 detection in X-ray images
  • Citing Article
  • November 2022

Methods

... To understand the gravity of the issue, contrast OT with another divergence, the Maximum Mean Discrepancy [34], whose sample complexity is dimension independent! Finally, there are limitations inherent to the vanilla OT formulation itself such as sensitivity to outliers [35], inability to incorporate context/structure in the optimal plan [36], strong reliance on the chosen ground cost [37], assuming a common ground cost [38] and the inability to handle measures (generalized, un-normalized distributions) of non-equivalent masses (normalization constants) [39]. Unsatisfied with these limitations, researchers have produced a diverse spectrum of OT formulations over the years with the potential to meet a broader range of applications and requirements. ...

A brief survey on Computational Gromov-Wasserstein distance
  • Citing Article
  • February 2022

Procedia Computer Science

... We can better understand the correlation between nodes and their labels when considering the semantic connections among the nodes through their neighborhoods. In Figure 2, nodes p1 and p2 have multi-label neighbors [36], so a path from p1 to p2 is determined as p1.2. This technique strengthens the communication between the nodes, and rich information is obtained from the paths. ...

Latent neighborhood-based heterogeneous graph representation
  • Citing Article
  • July 2022

Neural Networks