Distribution-Informed Neural Networks for Domain Adaptation Regression
Authors: Jun Wu, Jingrui He, Sheng Wang, Kaiyu Guan, Elizabeth Ainsworth
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The efficacy of our framework is also empirically verified on a variety of domain adaptation regression benchmarks. We experimentally investigate the performance of our DINO framework on a variety of domain adaptation regression benchmarks, and show its effectiveness over state-of-the-art baselines. The experiments demonstrate the effectiveness of our DINO framework over state-of-the-art baselines. |
| Researcher Affiliation | Academia | Jun Wu, Jingrui He, Sheng Wang, Kaiyu Guan, Elizabeth Ainsworth University of Illinois Urbana-Champaign {junwu3,jingrui,sheng12,kaiyug,ainswort}@illinois.edu |
| Pseudocode | Yes | As illustrated in Algorithm 1, we use all the training source (target) examples as the basis source (target) examples xr nr. (Section 4.2). Appendix A.1 contains "Algorithm 1: DINO-INIT Training and Prediction". |
| Open Source Code | Yes | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See the supplemental material. |
| Open Datasets | Yes | Following [10], we use two image data sets: d Sprites [40] and MPI3D [23]. In addition, we also use a plant phenotyping data set. (Section 5). These datasets are well-known and cited, implying public availability. |
| Dataset Splits | No | The paper describes the use of source and target labeled examples for training and refers to testing on unlabeled target examples. However, it does not explicitly define a separate validation dataset split with specific percentages or counts for hyperparameter tuning or early stopping during training. |
| Hardware Specification | Yes | Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Appendix A.11. All the experiments are implemented in PyTorch with a NVIDIA GeForce RTX 3090 GPU (Appendix A.11). |
| Software Dependencies | No | The paper mentions "implemented in PyTorch" (Appendix A.11) and that "The induced NNGP and neural tangent kernels induced can be estimated using the Neural Tangents package [41]" (Section 5), but it does not specify version numbers for PyTorch or Neural Tangents. |
| Experiment Setup | Yes | In the experiments, our algorithms are implemented using a L-layer (L = 6) fully-connected neural network with Re LU (see Appendix A.11 for more details). In addition, we set = 0.5 and µ = 0.1 for DINO-TRAIN (Section 5 - Implementations). The neural network structure is a 6-layer MLP with ReLU activation functions, and the width for each hidden layer is 1024. For training DINO-TRAIN, we use Adam optimizer with learning rate 0.001 and batch size 64 (Appendix A.11). |