Differentiable Learning Under Triage

Authors: Nastaran Okati, Abir De, Manuel Rodriguez

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on a wide variety of supervised learning tasks using synthetic and real data from two important applications content moderation and scientific discovery illustrate our theoretical results and show that the models and triage policies provided by our algorithm outperform those provided by several competitive baselines.
Researcher Affiliation Academia Nastaran Okati MPI for Software Systems nastaran@mpi-sws.org Abir De IIT Bombay abir@cse.iitb.ac.in Manuel Gomez-Rodriguez MPI for Software Systems manuelgr@mpi-sws.org
Pseudocode Yes Refer to Algorithm 1 for a pseudocode implementation of the overall gradient-based algorithm, which returns θT and γ.
Open Source Code Yes Our code and data are available at https://github.com/Networks-Learning/differentiable-learning-under-triage
Open Datasets Yes We use two publicly available datasets [33, 34], one from an application in content moderation and the other for scientific discovery. ... The Hatespeech dataset is publicly available under MIT license and the Galaxy zoo dataset is publicly available under Creative Commons Attribution-Noncommercial-No Derivative Works 2.0 license.
Dataset Splits Yes In each experiment, we used 60% samples for training, 20% for validation and 20% for testing.
Hardware Specification Yes All algorithms were implemented in Python 3.7 and ran on a V100 Nvidia Tesla GPU with 32GB of memory.
Software Dependencies No The paper mentions 'Python 3.7' but does not list multiple key software components with specific version numbers, nor a self-contained solver with a version, which would be needed for a 'Yes' classification based on the schema requirements.
Experiment Setup Yes In each experiment, we used 60% samples for training, 20% for validation and 20% for testing. ... we find the parameters of the predictive model mθt via stochastic gradient descent (SGD) [28], i.e., θ(j) t = θ(j 1) t α(j 1) θ L... where α(j) is the learning rate at iteration j.