Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
GLNCD: Graph-Level Novel Category Discovery
Authors: Bowen Deng, Lele Fu, Sheng Huang, Tianchi Liao, Jialong Chen, Zhang Tao, Chuan Chen
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on four GLNCD benchmark datasets demonstrate the effectiveness of Proto FGW-NCD. |
| Researcher Affiliation | Academia | Bowen Deng Sun Yat-sen University EMAIL Lele Fu Sun Yat-sen University Sheng Huang Sun Yat-sen University Tianchi Liao Sun Yat-sen University Jialong Chen Sun Yat-sen University Tao Zhang Sun Yat-sen University EMAIL Chuan Chen Sun Yat-sen University EMAIL |
| Pseudocode | Yes | Algorithm 1: POT BAPG Solver [39] for Fused Gromov-Wasserstein Distance (Forward) ... Algorithm 2: Our BAPG layer for Fused Gromov-Wasserstein Distance (Forward) |
| Open Source Code | No | We plan to release the code, datasets, and pre-trained models with sufficient instructions to faithfully reproduce our results. |
| Open Datasets | Yes | From these, we select three representative datasets spanning diverse domains, which, together with CIFAR10 (graph) [15], form the benchmark for GLNCD (Table 1). Bioinformatics: The ENZYMES dataset [55, 50] ... Program Analysis: The Mal Net-Tiny dataset [21] ... Social Networks: In the REDDIT12K dataset [71, 50] ... Computer Vision: The CIFAR10 (graph) dataset [15] is constructed from CIFAR10 (image) [38] in this way. |
| Dataset Splits | Yes | Table 7: The split information of four GLNCD datasets Dataset # train # test # all ENZYMES 420 120 600 Mal Net-Tiny 3500 1000 5000 REDDIT12K 8350 2386 11929 CIFAR10 35000 10000 60000 |
| Hardware Specification | Yes | All experiments are conducted on Ubuntu 22.04 server equipped with an RTX 4090 GPU and Intel Xeon Gold 6240C CPU. |
| Software Dependencies | Yes | The implementation is based on Pytorch2.5 [52] and Py G2.6 [17]. |
| Experiment Setup | Yes | Table 8: Hyperparameter values and search spaces of GLNCD methods Group Hyperparameter Value or Search Space Common Hyperparameters Optimization Learning rate [0.001, 0.005, 0.01, 0.05, 0.1] Dropout [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8] Weight decay [0.0, 5e-5, 1e-4, 5e-4, 1e-3, 5e-3, 1e-2] Cosine warmup steps [2, 5, 10] Batch size [64, 128, 256, 512] Max epochs [20, 50, 100, 300] Neural Network Arch. GNN encoder layer [2, 3, 4, 5, 6] Hidden dimension [32, 64, 128, 256] has_residual [False, True] has_ffn [False, True] Normalization [batchborm, None] Graph SSL Node droprate [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6] Temp. for contrastive loss [0.1, 0.3, 0.5, 0.7, 0.9, 1.1] GLNCD method Hyperparameters All baselines Encoder pooling readout [ mean , add , max ] Auto Novel Topk in RS [5, 10, 15] Rampup length [10, 50, 80, 150, 300] Rampup coefficient [1.0, 5.0, 25., 50.] NCL Labeled NCL loss weight [0.2, 1] Unlabeled NCL loss weight [0.2, 1] Queue length [200, 2000] Dual RS Memory bank length [256, 512, 1024] Proto FGW-NCD Epsilon [0.01, 0.05, 0.07, 0.11, 0.15, 0.19] Prototype node feature std. [0.5, 1.] # prototype graphs range(10, 130, 10) # prototype graph nodes 20 |