Data driven semi-supervised learning
Authors: Maria-Florina F. Balcan, Dravyansh Sharma
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section we evaluate the performance of our learning procedures when finding application-specific semi-supervised learning algorithms (i.e. graph parameters). Our experiments3 demonstrate that the best parameter for different applications varies greatly, and that the techniques presented in this paper can lead to large gains. We look at image classification based on standard pixel embedding. |
| Researcher Affiliation | Academia | Maria-Florina Balcan School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 ninamf@cs.cmu.edu Dravyansh Sharma Department of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 dravyans@cs.cmu.edu |
| Pseudocode | Yes | Algorithm 1 Data-driven Graph-based SSL and Algorithm 2 Efficient Data-driven Graph-based SSL are provided in the paper. |
| Open Source Code | Yes | Code: https://drive.google.com/drive/folders/1Iq Iw2Mp23W35UUwlz1hy24Eba5s Pp VH_ |
| Open Datasets | Yes | We use three popular benchmark datasets MNIST [Le Cun et al., 1998], Omniglot [Lake et al., 2015] and CIFAR-10 [Szegedy et al., 2015]. |
| Dataset Splits | No | The paper describes a sampling method for creating semi-supervised learning instances (e.g., "sampling 100 random examples and further sampling L random examples from the subset for labeling") but does not provide traditional training/test/validation dataset splits (e.g., percentages or fixed counts for global dataset partitions). |
| Hardware Specification | No | The paper does not explicitly mention the specific hardware (e.g., GPU/CPU models, memory) used for running the experiments in the main text. The information in the checklist is meta-information and not part of the paper's content. |
| Software Dependencies | No | The paper does not provide specific software names with version numbers for dependencies used in the experiments. |
| Experiment Setup | Yes | Setup: ... n = 100 ... L = 10 for MNIST, while L = 20 for Omniglot and CIFAR-10. ... determine data-specific good values for σ, when predictions are made by optimizing the harmonic objective (Table 1). ... σ [0, 10]. ... average over 50 iterations for learning from 50 problem instances each (T = 50, Figure 3). |