D2R2: Diffusion-based Representation with Random Distance Matching for Tabular Few-shot Learning

Authors: Ruoxue Liu, Linjiajie Fang, Wenjia Wang, Bingyi Jing

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments reveal the significant efficacy of D2R2 across various tabular few-shot learning benchmarks, demonstrating its state-of-the-art performance in this field.
Researcher Affiliation Academia Ruoxue Liu HKUST rliuaj@connect.ust.hk Linjiajie Fang HKUST lfangad@connect.ust.hk Wenjia Wang HKUST (GZ) and HKUST wenjiawang@ust.hk Bing-Yi Jing SUSTech jingby@sustech.edu.cn
Pseudocode Yes A Algorithm Algorithm 1 Training D2R2 ... Algorithm 2 Pseudo-label Validation
Open Source Code Yes Code available at https://github.com/Carol-cloud-project/D2R2
Open Datasets Yes We select nine datasets from the Open ML-CC18 benchmark [3, 6] to validate the performance of D2R2.
Dataset Splits Yes For all the datasets, we randomly split 80% of the data for training and the remaining 20% for testing. ... Additionally, 20% of the training data is utilized for validation and hyperparameter tuning.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No Table 4 mentions 'Optimizer Adam [23]' but does not specify a software version for Adam or any other libraries like Python, PyTorch, or TensorFlow.
Experiment Setup Yes D.3 Hyperparameter Details. We employ fully connected layers with identical layer counts and hidden dimensions to model the diffusion noise and the embedding function. The training procedure for the diffusion model follows the same approach as in DDPM [17]. The configuration for the diffusion model and neural networks (NN) are as following: Table 4: Model configurations. Parameter Setting Description T 10 Diffusion timesteps. Beta_t vp [45] Noise schedule. Hidden dims 512 Dimension of dense layers. Layers 3 Number of dense layers. Optimizer Adam [23] Batch Size 256 Learning Rate 0.0003. Table 5: Hyperparameter search range Parameter Range Description alpha linespace(0.1, 0.1, 5) RDM weight. dz {5, 10, 20, 40, 80} Embedding dimension tau linespace(0.1, 0.1, 2) Temperature in equ. (6)