Multi-Stage Multi-Task Learning with Reduced Rank
Authors: Lei Han, Yu Zhang
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical studies on synthetic and real-world datasets demonstrate the effectiveness of the RAMUSA method in comparison with the state-of-the-art methods. In this section, we conduct empirical experiments on one synthetic dataset and five real-world datasets to study the proposed RAMUSA method. |
| Researcher Affiliation | Academia | 1Department of Statistics, Rutgers University 2Department of Computer Science and Engineering, Hong Kong University of Science and Technology |
| Pseudocode | Yes | Algorithm 1 Multi-Stage Algorithm for the RAMUSA Model |
| Open Source Code | No | No explicit statement about open-sourcing the code for the methodology, nor a link to a code repository, was found. |
| Open Datasets | Yes | School Data: the objective is to predict the student exam scores in different schools. Tasks correspond to schools, features are attributes for describing students, and each task has a different number of samples corresponding to students. We randomly select 10%, 20% and 30% of the samples from each task as the training set and the rest as the test set; SARCOS Data: the problem is an inverse dynamics prediction problem for a seven degrees-of-freedom anthropomorphic robot arm, which needs to map from the feature space to seven joint torques. We randomly select 100, 200 and 300 samples to form the training set and randomly select 5000 samples to form the test set; Microarray Data: this is a gene expression data set related to isoprenoid biosynthesis. The tasks are finding the crosstalks from the mevalonate genes to the plastidial genes. We randomly select 20% and 40% of the samples as the training set and use the rest for testing; Traffic Data: this is to find the casual relationships from the entries to the exits in a highway traffic network, where each exit corresponds to one task and the information collected in entries is considered as the features shared by all the tasks. The settings are the same as those in the Microarray data; Handwritten Letter Data: the goal is to discriminate between 7 pairs of letters, i.e. c/e, g/y, m/n, a/g, a/o, f/t and h/n. The features are pixel values of the handwritten letter. We randomly choose 10% and 20% of the samples as the training sets and the rest as the test set. |
| Dataset Splits | Yes | We generate 50 and 200 samples for training and testing separately and use another 200 samples as a validation set to select the regularization parameters and hyperparameters in all the compared methods including the parameter τ in the RAMUSA method. |
| Hardware Specification | No | No specific hardware details such as GPU/CPU models, memory, or specific computing environments were mentioned for running the experiments. |
| Software Dependencies | No | The paper mentions methods like FISTA, Lasso, and MTFL but does not list specific software packages or libraries with version numbers used for implementation or experimentation. |
| Experiment Setup | Yes | We generate 50 and 200 samples for training and testing separately and use another 200 samples as a validation set to select the regularization parameters and hyperparameters in all the compared methods including the parameter τ in the RAMUSA method. For the parameter τ in the RAMUSA method, we choose it in a candidate set [10 3, 10 2, , 103] via 5-fold cross validation. |