Efficient Continuous Pareto Exploration in Multi-Task Learning

Authors: Pingchuan Ma, Tao Du, Wojciech Matusik

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

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically evaluate our method on fve datasets with various size and model complexity, ranging from Mul ti MNIST (Sabour et al., 2017) that consists of 60k images and requires a network classifer with only 20k parameters, to UTKFace (Zhang et al., 2017), an image dataset with 3 objectives and a modern network structure with millions of parameters.
Researcher Affiliation Academia Pingchuan Ma * 1 Tao Du * 1 Wojciech Matusik 1 1MIT CSAIL. Correspondence to: Pingchuan Ma <pcma@csail.mit.edu>.
Pseudocode Yes Algorithm 1 Effcient Pareto Set Exploration
Open Source Code Yes The code and data are available online1. 1https://github.com/mit-gfx/Continuous Pareto MTL
Open Datasets Yes Multi MNIST (Sabour et al., 2017) and its two variants Fashion MNIST (Xiao et al., 2017) and Multi Fashion M- NIST, which are medium-sized datasets with two classi fcation tasks; 2) UCI Census-Income (Kohavi, 1996), a medium-sized demographic dataset with three binary pre diction tasks; 3) UTKFace (Zhang et al., 2017), a large dataset of face images.
Dataset Splits No The paper mentions using well-known datasets and a subset of Multi MNIST, but does not explicitly state the specific training, validation, or test data splits (e.g., percentages or exact counts) within the main body of the paper.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using the minimal residual method (MINRES) but does not provide specific version numbers for software dependencies or libraries used for implementation (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We discussed in more detail two crucial hyperparameters (k and s) and reported the ablation study in Section 6. We started with a random Pareto station ary point x returned by Pareto MTL, followed by running Algorithm 1 with fxed parameters K = 1 and N = 5 on Multi MNIST and its two variants.