Multi-task Learning by Leveraging the Semantic Information

Authors: Fan Zhou, Brahim Chaib-draa, Boyu Wang11088-11096

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To confirm the effectiveness of the proposed method, we first compare the algorithm with several baselines on some benchmarks and then test the algorithms under label space shift conditions. Empirical results demonstrate that the proposed method could outperform most baselines and achieve state-of-the-art performance, particularly showing the benefits under the label shift conditions.
Researcher Affiliation Academia Fan Zhou1, Brahim Chaib-draa1, Boyu Wang2,3* 1 Universit e Laval, Quebec City, QC, G1V 0A6, Canada. 2 University of Western Ontario, London, ON N6A 5B7, Canada 3 Vector Institute, Toronto, ON M5G 1M1, Canada fan.zhou.1@ulaval.ca, chaib@ift.ulaval.ca, bwang@csd.uwo.ca
Pseudocode Yes Algorithm 1 The Global Semantic Matching Method Algorithm 2 The proposed Semantic Multi-task learning algorithm
Open Source Code No The paper does not contain an explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We examined the proposed approach comparing with several baselines on Digits, PACS (Li et al. 2017), Office31 (Saenko et al. 2010), Office-Caltech (Gong et al. 2012) and Office-home (Venkateswara et al. 2017) dataset.
Dataset Splits Yes For the Digits benchmark, we evaluate the algorithms on MNIST (M), MNIST-M (M-M) and SVHN (S) simultaneously. ...we follow the evaluation protocol of (Shui et al. 2019) by randomly selecting 3k, 6k and 8k instances of the training dataset and choose 1k dataset as validation set while testing one the full test set.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments.
Software Dependencies No The paper mentions "Py Torch (Paszke et al. 2019)" but does not specify a version number for it or any other software dependencies.
Experiment Setup Yes The model is trained for 50 epochs while the initial learning rate is set by 1 × 10−3 and is decayed 5% for every 5 epochs. ... We adopt the Adam optimizer with initial learning rate 2 × 10−4 and decayed 5% every 5 epochs while totally training for 80 epochs. For stable training, we also enable the weight-decay in Adam optimizer to enforce a L2 regularization.