Improved Active Multi-Task Representation Learning via Lasso
Authors: Yiping Wang, Yifang Chen, Kevin Jamieson, Simon Shaolei Du
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we provide experiments on realworld computer vision datasets to illustrate the effectiveness of our proposed method. Finally, we empirically show the effectiveness of our algorithms. If we denote the practical algorithm of (Chen et al., 2022) by L2-A-MTRL, we show that our proposed L1-A-MTRL algorithm achieves 0.54% higher average accuracy on MNIST-C relative to L2-AMTRL (92.6%), which confirms our theoretical results. |
| Researcher Affiliation | Academia | 1College of Computer Science and Technology, Zhejiang University 2Paul G. Allen School of Computer Science & Engineering, University of Washington. |
| Pseudocode | Yes | Algorithm 1 L1-A-MTRL Method |
| Open Source Code | No | The paper does not provide an explicit statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | Datasets. The MNIST-C dataset is a comprehensive suite of 16 corruptions applied to the MNIST test set. Like in (Chen et al., 2022), we divide each corruption-related sub-dataset into 10 tasks according to their labels ranging from 0 9 and thus get 160 separate new tasks denoted by {corruption type} {label} . For instance, brightness 0 denotes the data corrupted by brightness noise and relabeled to 1/0 based on whether the data corresponds to number 0 or not. And once we choose 1 task called {type A} {label B} for the target task, the other 150 tasks that don t contain type A corruption will be chosen as source tasks. |
| Dataset Splits | No | The paper mentions using the MNIST-C dataset and sampling data from target and source tasks but does not provide specific details on how the data was split into training, validation, and test sets with percentages or counts. |
| 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 does not provide specific software details, such as library names with version numbers, needed to replicate the experiment. |
| Experiment Setup | Yes | We set d = 28 28, k = 50 and there are T = 150 source tasks in total. Here we set N = 100. We sample 500 data from the target task, while at the final stage, we sample around 30000 to 40000 data from the source tasks. |