Embedded Files
Federated Learning(FL) is a Google-proposed paradigm that implements a distributed machine learning approach in which, rather than learning from centralized data or from the data which is distributed among machines in a balanced and IID fashion, they formulate mechanisms to learn from subsets of data distributed in non IID fashion and then aggregate the learnings, helping models to run locally and reliably on mobile devices while keeping their training data private to each node and never exchanged during the learning process. The current research on this is going on Kharagpur Learning, Imaging & Visualization (KLIV) research group under the Supervision of Dr. Debdoot Sheet, Assistant Professor, Department of Electrical Engineering, Indian Institute of Technology, Kharagpur.
Our Work (Ongoing)
Our current work is on the process of building a state of the art Federated Learning architecture addressing all the current challenges and develop a full scale federated learning environment from the state of art centralized FL to a fully decentralized industry ready FL with notation of privacy, verification of the local models along with an incentive mechanism incorporated in the architecture.
We are developing our own framework where different institutions can collaborate in the federated learning process in a fast and effective way focusing on some of the loop holes present in the current frameworks. Along with this we are incorporating some more features in the current state of the art federated learning framework which will make the architecture more robust and secure against malicious clients. We will be shortly releasing one introductory video of the framework.
Fed_Sock.pdfDeveloping a Industry Ready Federated Learning Framework (Sample Instruction Flow)
Dataset_Distribution.pptxDataset Distribution
Differential Privacy.pptxDifferential Privacy in Federated Learning
Socket Programing.pptxSample Working Flow of the framework
Need For Fed_Prox.pptxImportance of FedProx Algorithm
Federated Dropout.pptxFederated Dropout (Overview)
Some Preliminary Results
Current State of the Art Papers
Communication- Efficient Learning of Deep Networks from Decentralized Data. This is the original paper where Google on federated. arxiv.org/abs/1602.05629
Federated Optimization in Heterogeneous Network: Tackles the heterogeneity in federated networks arxiv.org/abs/1812.06127
Federated Learning With Matched Averaging: Here, authors proposed another method to aggregate the model parameters which they claim to outperform the state of the start Federated Average Algorithm which uses Bayesian nonparametric methods to adapt to heterogeneity in the data. arxiv.org/abs/2002.06440
Federated Learning for Site Aware Chest Radiograph Screening ieeexplore.ieee.org/document/9433876
FedVision (Link)
Federated Visual Classification with Real-World-Data Distribution (Link)
Federated Learning in Vehicular Networks: (Link)
The Future of Digital Health with Federated Learning: (Link)
Blockchained On-Device Federated Learning (Link)
Frameworks and Codes: Online videos:
PySyft by OpenMined: (Link ) 1. Lecture on Distributed Federated Learning by Dr. Debdoot Sheet: YouTube-Link
Tensor Flow Federated: www.tensorflow.org/federated 2. Udacity : Link (Free)
Flower framework : flower.dev/ 3. Udemy Link (Paid)
Refer to this playlist for more details YouTube-Link 4. Foundations of Private Computation Link (Free)
Code: Simple implementation of Fed-Avg Algorithm in GitHub-Link
Tensorflow Privacy (Link)
Pytorch Opacus (Link)
For more frameworks: Link
Page updated
Google Sites
Report abuse