Adversarially Robust Multi-task Representation Learning

Authors: Austin Watkins, Thanh Nguyen-Tang, Enayat Ullah, Raman Arora

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper, we study this problem from a theoretical perspective, building on prior work on this topic [13, 26, 12, 42]. We focus on transfer learning via learning a representation that provides a method for sharing knowledge between different, albeit related, tasks [7, 45, 14, 29, 49]. A common approach to achieve this has been termed multi-task representation learning (MTRL) [9, 24, 11]. In practice, the class of representations is a complex model like a deep neural network, and the predictors trained on top of them are simple linear models [7, 14]. Such a paradigm offers hope that we can pool our data, potentially providing a substantial benefit if performed well.
Researcher Affiliation Collaboration Austin Watkins Johns Hopkins University Baltimore, MD 21218 awatki29@jhu.edu Thanh Nguyen-Tang Johns Hopkins University Baltimore, MD 21218 nguyent@cs.jhu.edu Enayat Ullah Meta enayat@meta.com Raman Arora Johns Hopkins University Baltimore, MD 21218 arora@cs.jhu.edu
Pseudocode Yes Algorithm 1 (Two-stage adversarial MTRL).
Open Source Code No The paper is theoretical and does not mention providing open-source code for the described methodology. It does not contain any links to code repositories or statements about code availability.
Open Datasets No The paper is purely theoretical and does not involve experimental data or datasets. Therefore, it does not specify public availability of datasets used for training.
Dataset Splits No The paper is purely theoretical and does not involve experimental data or datasets. Therefore, it does not specify training/test/validation dataset splits.
Hardware Specification No The paper is purely theoretical and does not describe any experiments or the hardware used to run them.
Software Dependencies No The paper is purely theoretical and does not describe any experiments or the software dependencies with specific versions needed for reproduction.
Experiment Setup No The paper is purely theoretical and does not describe any experimental setup details such as hyperparameters or training configurations.