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..
Understanding Bias Terms in Neural Representations
Authors: Weixiang Zhang, Boxi Li, Shuzhao Xie, Chengwei Ren, Yuan Xue, Zhi Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To validate our analysis from Sec. 2 regarding bias terms impact in INRs, we conduct empirical studies under various bias-related configurations. Following prior work [41, 27, 61, 48], we implement 2D image fitting tasks on the Div2K dataset [1] using a 3 256 SIREN network (detailed settings and alternative backbones are provided in the appendix)., 4 Experiments, Quantitative Results. The quantitative results are presented in Tab. 1., 4.3 Ablation Study: Additional Verification for Spatial Aliasing |
| Researcher Affiliation | Academia | Weixiang Zhang, Boxi Li, Shuzhao Xie, Chengwei Ren, Yuan Xue, Zhi Wang Shenzhen International Graduate School, Tsinghua University EMAIL, EMAIL EMAIL |
| Pseudocode | No | The paper describes methods and procedures in paragraph text, but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks or figures. |
| Open Source Code | Yes | Our code is available at this link. |
| Open Datasets | Yes | We conduct 2D image fitting tasks on the Div2K dataset [1], achieves only approximately 60% accuracy on the CIFAR-10 dataset., Dataset of Neural Representations. Beyond standard image datasets, establishing a large-scale dataset of neural representations is crucial. While Implicit-Zoo [29] provides CIFAR-10 in neural representation form |
| Dataset Splits | No | We conduct 2D image fitting tasks on the DIV2K dataset [1], we re-train the CIFAR-10 neural representation dataset using a 1 64 SIREN network trained for 1000 epochs, maintaining alignment with Implicit Zoo [29] parameters except for the bias initialization scheme., Tab. 8 shows the classification results for CIFAR-100 dataset [24]. The paper refers to standard datasets but does not explicitly state the train/validation/test splits, percentages, or sample counts used for reproduction within the provided text. |
| Hardware Specification | Yes | All experiments in this section were performed using the Py Torch framework [36] on NVIDIA RTX 3090 GPU with 24.58 GB VRAM. |
| Software Dependencies | No | All experiments in this section were performed using the Py Torch framework [36]. While PyTorch is mentioned, a specific version number is not provided. |
| Experiment Setup | Yes | Following prior work [41, 27, 61, 48], we conduct 2D image fitting tasks on the DIV2K dataset [1] (additional datasets are detailed in supplementary materials). We employ a 3 256 MLP with SIREN [45] and FINER [27] architectures, setting the total number of iterations T to 5000., The downstream network of Feat-Bias consists of a lightweight 3 256 MLP classifier, trained for 1000 iterations using a cosine scheduler with a learning rate of 1e 3., using a 1 64 SIREN network trained for 1000 epochs |