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..
GSRF: Complex-Valued 3D Gaussian Splatting for Efficient Radio-Frequency Data Synthesis
Authors: Kang Yang, Gaofeng Dong, Sijie Ji, Wan Du, Mani Srivastava
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments Our method is implemented in Py Torch with CUDA. Further implementation details and hyperparameter settings are provided in Appendix E, and additional experiments are presented in Appendix F. We evaluate GSRF across three RF technologies for various RF data synthesis tasks: (i) Radio Frequency Identification (RFID) for spatial spectrum synthesis, (ii) Bluetooth Low Energy (BLE) for real-valued received signal strength indicator (RSSI) synthesis, (iii) 5G Cellular Network for complex-valued channel state information (CSI) [38] synthesis. |
| Researcher Affiliation | Academia | Kang Yang1 Gaofeng Dong1 Sijie Ji1,2 Wan Du3B Mani Srivastava1 1University of California, Los Angeles 2California Institute of Technology 3University of California, Merced EMAIL EMAIL EMAIL B EMAIL |
| Pseudocode | Yes | Algorithm 1: Forward CUDA Kernel for Ray Tracing Algorithm Input: w, h: numbers of rays in azimuth and elevation Input: M, C: means & covariances of all Gaussians Input: E, A: radiances & transmittances of all Gaussians Input: L: positions of receiver and transmitter Output: O: received signals for all rays |
| Open Source Code | Yes | We release our code at this Git Hub repository. |
| Open Datasets | Yes | The publicly released RFID dataset from Ne RF2 [12], collected in real-world indoor environments, is employed. |
| Dataset Splits | Yes | The publicly released RFID dataset from Ne RF2 [12]... The dataset is randomly split by default into 70% for training and 30% for testing. |
| Hardware Specification | Yes | Training time is measured by running each method on a computer equipped with Ge Force RTX 3080Ti GPU. |
| Software Dependencies | No | Our method is implemented in Py Torch with CUDA. |
| Experiment Setup | Yes | Table 2: Hyperparameter settings. ... The attributes of all 3D Gaussians are updated using SGD [47]:... The learning rates are set as follows: ηρ = 0.01 for attenuation, ηψ = 0.0025 for emission, ηS = 0.01 for the scaling matrix, and ηR = 0.005 for the rotation matrix. The learning rate for the mean, ηµ, starts at 0.00016 and decreases exponentially to 1.6 10 6 over 30,000 iterations. |