June Evaluation

Goal attainment

  • Finish most part of the reproduction. CloudyFL: A Cloudlet-Based Federated Learning Framework for Sensing User Behavior Using Wearable Devices

    • Do experiments on IID data.

      • Centralized softmax regression + MLP($84%$, $98%$)

      • FedAvg softmax regression + MLP ($82%$, $98%$)

    • non-IID data (give each clients different labels, $40$ rounds)

      • FedAvg + softmax ($52%$)

      • FedAvg + MLP ($60%$)

      • FedAvg + LSTM ($79%$)

      • FedRS + LSTM (also around $79%$)

    • Write $3$ simulation files:

      • First one: Do centralized and FedAvg, just mainly follow how the Author write, but convert torch to tensorflow

      • Second one: Rewrite getData function, which becomes time series and can apply LSTM/RNN

      • Third one: Rewrite all codes, re-split the data, rewrite softmax to be a restrict softmax in the last layer. In this stage, ML-models are easy to apply.

  • Read some papers.

    • CloudyFL: A Cloudlet-Based Federated Learning Framework for Sensing User Behavior Using Wearable Devices

    • Personalized Federated Recommendation via Joint Representation Learning, User Clustering, and Model Adaptation

    • Communication-Efficient Learning of Deep Networks from Decentralized Data

    • FedStack: Personalized activity monitoring using stacked federated learning

Skills development

  • Learned ML models

    • LR(softmax R) + SVM + MLP + CNN + RNN + LSTM

    • Knowledges are learned from

      • Dive into Deep Learning (https://d2l.ai)

      • CS231N

      • A book

      • Ask others

  • Learned Federated Learning

    • FedAvg

    • FedRS

  • Learned DP (here means differential privacy)

PreviousSubwork_TianjunNextSubwork_Aicha

Last updated