Distributed Primal-Dual Optimization for Online Multi-Task Learning

Authors: Peng Yang, Ping Li6631-6638

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results confirm that the proposed model is highly effective on various real-world datasets. Empirical experiments are conducted to evaluate the algorithms on three datasets used in previous work (Zhang et al. 2018).
Researcher Affiliation Industry Peng Yang, Ping Li Cognitive Computing Lab Baidu Research 10900 NE 8th ST, Bellevue WA, 98004, USA {pengyang01, liping11}@baidu.com
Pseudocode Yes Algorithm 1 DROM: Distributed Primal-dual optimization for Online MTL
Open Source Code No The paper states that implementations for baselines are provided in 'Supporting Materials' but does not explicitly state that the source code for their proposed method (DROM/DROM-D) is open-source or provide a link.
Open Datasets Yes Spam Email2 contains 7,068 emails collected from mailboxes of 4 users (i.e., 4 tasks). (Footnote 2: http://labs-repos.iit.demokritos.gr/skel/i-config/) MHC-I3, a bio-marker dataset, contains 18,664 peptide sequences for 12 MHC-I molecules (i.e., 12 tasks). (Footnote 3: http://web.cs.iastate.edu/~honar/ailab/) Each Movie4 is a movie recommendation dataset where 72,916 users rate a subset of 1,628 movies. (Footnote 4: http://goldberg.berkeley.edu/jester-data/)
Dataset Splits No The paper describes an online learning setting and evaluates performance using cumulative error rate over online data but does not provide specific train/validation/test dataset splits or their sizes.
Hardware Specification No The paper reports runtime in seconds but does not provide any specific hardware details such as GPU or CPU models used for the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes For both methods, we simply set λ = 1, ρ = 1 to avoid overfitting, and tune p (0, 1) with ξ = 1 to deal with adversarial noise. When ηt = 1/ t the following regret is hold (for DROM) and When ηt = 1/ t/τ , the regret holds (for DROM-D).