README.md

FedSeIT: Federated Selective Inter-Client Transfer

This code repository contains the implementations of: (a) "Federated Selective Inter-client Transfer (FedSeIT)" framework, and (b) "Selective Inter-client Transfer (SIT)" method proposed in the paper titled: "Federated Continual Learning for Text Classification via Selective Inter-client Transfer" accepted at EMNLP2022 Conference in "Findings in EMNLP" track.

Highlights of the FedSeIT paper

1. The authors propose a novel framework FedSeIT which performs domain alignment of foreign clients' task-specific parameters to minimize inter-client interference.
2. The authors propose a novel task selection method SIT which efficiently selects relevant parameters from foreign clients to maximize inter-client knowledge transfer.
3. It is the first work that applies FCL method in NLP domain.
4. State-of-the-art performance on Text Classification task over 5 NLP datasets from diverse domains.

Environment Setup

Download 300-dimensional "glove.840B.300d.txt" GloVe embeddings from this link and place it in Resources/ directory.

> python3.8 -m venv venv
> source venv/bin/activate
> pip3 install -r requirements.txt

FCL Data Generation and Preprocessing

This bash script below needs to be run only once, to create tasks from the dataset in a random order (with fixed random seed for reproducibility). Here we have already provided the preprocessed data for Reuters8 (R8) dataset. However, the bash scripts are also provided for TMN and TREC6 datasets.

> cd Config/r8/
> bash data.sh

Run CNN Text Classification model with FedWeIT framework

All configurable parameters for model training in FCL setting can be found in FedWeIT/parser.py file.

> cd Config/r8/
> bash FedWeIT.sh

Run CNN Text Classification model with FedSeIT framework

> cd Config/r8/
> bash FedSeIT.sh

How to cite this work

@inproceedings{chaudhary-etal-2022-federated,
    title = "Federated Continual Learning for Text Classification via Selective Inter-client Transfer",
    author = {Chaudhary, Yatin  and
      Rai, Pranav  and
      Schubert, Matthias  and
      Sch{\"u}tze, Hinrich  and
      Gupta, Pankaj},
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2022",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.findings-emnlp.353",
    pages = "4789--4799",
    abstract = "In this work, we combine the two paradigms: Federated Learning (FL) and Continual Learning (CL) for text classification task in cloud-edge continuum. The objective of Federated Continual Learning (FCL) is to improve deep learning models over life time at each client by (relevant and efficient) knowledge transfer without sharing data. Here, we address challenges in minimizing inter-client interference while knowledge sharing due to heterogeneous tasks across clients in FCL setup. In doing so, we propose a novel framework, Federated Selective Inter-client Transfer (FedSeIT) which selectively combines model parameters of foreign clients. To further maximize knowledge transfer, we assess domain overlap and select informative tasks from the sequence of historical tasks at each foreign client while preserving privacy. Evaluating against the baselines, we show improved performance, a gain of (average) 12.4{\%} in text classification over a sequence of tasks using five datasets from diverse domains. To the best of our knowledge, this is the first work that applies FCL to NLP.",
}

Existing FCL Work: FedWeIT

This code is based on the implementation of FedWeIT proposed by:

Yoon, Jaehong, Wonyong Jeong, Giwoong Lee, Eunho Yang, and Sung Ju Hwang. "Federated continual learning with weighted inter-client transfer." In International Conference on Machine Learning, pp. 12073-12086. PMLR, 2021.

Here, we have made a large number of significant contributions to the FedWeIT code to implement our proposed FedSeIT framework and SIT method, as cited in our research literature.