Active Ordinal Querying for Tuplewise Similarity Learning
Authors: Gregory Canal, Stefano Fenu, Christopher Rozell3332-3340
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that the performance of Info Tuple at various tuple sizes exceeds that of the state-of-the-art adaptive triplet selection method on synthetic tests and new human response datasets, and empirically demonstrate the significant gains in efficiency and query consistency achieved by querying larger tuples instead of triplets. |
| Researcher Affiliation | Academia | School of Electrical and Computer Engineering 2School of Interactive Computing Georgia Institute of Technology, Atlanta, GA {gregory.canal, sfenu3, crozell}@gatech.edu |
| Pseudocode | Yes | Algorithm 1 Info Tuple-k |
| Open Source Code | Yes | Code available at https://github.com/siplab-gt/infotuple |
| Open Datasets | Yes | Drawing 3000 food images from the Food-10k dataset (Wilber et al. 2015) |
| Dataset Splits | Yes | The validation set for this search was an additional 500 heldout triplets from the Food10k dataset. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models (e.g., NVIDIA A100), CPU models (e.g., Intel Xeon E5), or specific cloud instance types used for running the experiments. It mentions computational complexity abstractly (e.g., O(Nfk^2)) but not the physical machines. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9, CUDA 11.1). |
| Experiment Setup | Yes | For each of the human-subject experiments, μ was set to 0.1 and d was set to 4 per the hyperparameter search shown in Figure 5. In the synthetic experiments provided, μ was set to 0.5 and d was set to 2 to match the dimensionality of the generating distribution. A heuristic was used to pick a number of samples for the Monte Carlo estimation of the mutual information, with N 10 samples being used in practice. Turk subjects were presented with queries in batches of 25, with one repeated tuple across the batch as a test for validity. |