Adaptive Smoothed Online Multi-Task Learning

Authors: Keerthiram Murugesan, Hanxiao Liu, Jaime Carbonell, Yiming Yang

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments over three multitask learning benchmark datasets show advantageous performance of the proposed approach over several state-of-the-art online multi-task learning baselines. We evaluate the performance of our algorithm under batch and online settings. All reported results in this section are averaged over 30 random runs or permutations of the training data.
Researcher Affiliation Academia Keerthiram Murugesan Carnegie Mellon University kmuruges@cs.cmu.edu Hanxiao Liu Carnegie Mellon University hanxiaol@cs.cmu.edu Jaime Carbonell Carnegie Mellon University jgc@cs.cmu.edu Yiming Yang Carnegie Mellon University yiming@cs.cmu.edu
Pseudocode Yes Algorithm 1: Batch Algorithm (SMTL-e) Algorithm 2: Online Algorithm (OSMTL-e)
Open Source Code No The paper does not provide a clear statement or link for the availability of its source code.
Open Datasets Yes Landmine Detection3 consists of 19 tasks collected from different landmine fields. [...] 3http://www.ee.duke.edu/~lcarin/Landmine_Data.zip Spam Detection4 We use the dataset obtained from ECML PAKDD 2006 Discovery challenge for the spam detection task. [...] 4http://ecmlpkdd2006.org/challenge.html Sentiment Analysis5 We evaluated our algorithm on product reviews from amazon. [...] 5http://www.cs.jhu.edu/~mdredze/datasets/sentiment
Dataset Splits Yes Unless otherwise specified, all model parameters are chosen via 5-fold cross validation.
Hardware Specification No The paper does not provide specific hardware details (such as GPU or CPU models) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., library or solver names with versions) needed to replicate the experiment.
Experiment Setup Yes Corollary 2 provides a principled way to set hyperparameters to achieve the sub-linear regret bound. Specifically, recall α quantifies the self-concentration of each task. Therefore, α = T T implies for large horizon it would be less necessary to rely on other tasks as available supervision for the task itself is already plenty; C = 1+ T T T 0 suggests diminishing learning rate over the horizon length. Unless otherwise specified, all model parameters are chosen via 5-fold cross validation.