Online Multi-Task Learning via Sparse Dictionary Optimization
Authors: Paul Ruvolo, Eric Eaton
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluated four different algorithms for lifelong learning: a) ELLA, as defined in (Ruvolo and Eaton 2013), b) ELLA-SVD, c) ELLA Incremental, and d) ELLA Dual Update. Each algorithm was tested on four multi-task data sets: Synthetic Regression Tasks, Student Exam Score Prediction, Land Mine Detection, Facial Expression Recognition. The results of our evaluation are given in Figure 2. |
| Researcher Affiliation | Academia | Paul Ruvolo Franklin W. Olin College of Engineering Department of Engineering paul.ruvolo@olin.edu Eric Eaton University of Pennsylvania Computer and Information Science Department eeaton@cis.upenn.edu |
| Pseudocode | Yes | Algorithm 1 K-SVD (Aharon et al. 2006) and Algorithm 2 MTL-SVD |
| Open Source Code | No | The paper does not contain an unambiguous statement about releasing source code for the described methodology or a direct link to a code repository. |
| Open Datasets | Yes | London Schools data set... We use the same feature encoding as used by Kumar & Daum e (2012), where four school-specific and three student-specific categorical variables are encoded as a collection of binary features. Land Mine Detection... The data set contains a total of 14,820 data instances divided into 29 different geographical regions. We treat each geographical region as a different task. (Xue et al. 2007) Facial Expression Recognition... This data set is from a recent facial expression recognition challenge (Valstar et al. 2011). We use the same feature encoding as Ruvolo & Eaton (2013) |
| Dataset Splits | No | For each task, the training data was divided into both a training and a held-out test set (with 50% of the data designated for each). While a train/test split is mentioned, a separate validation split is not specified. |
| Hardware Specification | No | The paper does not provide specific hardware details (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 using 'linear or logistic regression' as the base learner and solving a problem 'using the Lasso (Tibshirani 1996)', but it does not provide specific software names with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9'). |
| Experiment Setup | Yes | The λ and k parameters were independently selected for each method via a gridsearch over all combinations of λ {e−5, . . . , e5} and k {1, . . . , 10}; µ was set to e−5 for all algorithms. |