Task Understanding from Confusing Multi-task Data
Authors: Xin Su, Yizhou Jiang, Shangqi Guo, Feng Chen
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We constructed a series of benchmarks for the CSL problem which includes function regression tasks and pattern recognition tasks. |
| Researcher Affiliation | Academia | 1Department of Automation, Tsinghua University, Beijing, 100084, China 2LSBDPA Beijing Key Laboratory, Beijing, 100084. China 3Beijing Innovation Center for Future Chip, Beijing, 100084, China. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks clearly labeled as such. |
| Open Source Code | No | The paper does not provide any concrete access information (e.g., repository link, explicit statement of code release, or mention of code in supplementary materials) for the described methodology. |
| Open Datasets | Yes | Colorful-Mnist: We extended the MNIST dataset (Le Cun et al., 1998) by adding random color in all images and obtained 0-9 digital images in 8 different colors, shown in Figure 4(a). Kaggle Fashion Product: We used a fashion dataset on Kaggle(Arslan et al., 2019) to construct a CSL recognition tasks for general objects, as shown in Figure 4(b). |
| Dataset Splits | No | The paper mentions training and testing but does not explicitly provide specific details about validation dataset splits (percentages, sample counts, or citations to predefined validation splits). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions the use of neural networks but does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, PyTorch 1.9) needed to replicate the experiment. |
| Experiment Setup | Yes | Without extra constraints, although a normal CSL result is a reasonable continuous function solution, it differs from the ground-truth. This is the difficulty of the CSL problem, mentioned in Section 3.3, in that multiple solutions could lead the learning risk converging towards zero. Therefore, we introduced a few-shot (5-shot) warm-up to determine the initialization of the neural network. |