pguijas / p2pfl

P2P Federated Learning (p2pfl) is a decentralized federated learning library, it allows creating basic federated learning systems on p2p networks using gossip protocols.

Hi!👋 My name is Pedro Guijas.

• 📚 Graduated in Computer Engineering and currently coursing a Master in Artificial Intelligence at UDC (University of Coruña).
• 🖥️ See my portfolio at TO-DO.
• 🚀 I'm currently working on p2pfl
• 🧠 I'm learning Deep Learning Generative Models.
• ⚡ Despite my main professional focus is Machine Learning, I also love software development.
• ✉️ You can contact me at [email protected].

My Skill Set

Connect with me

Github Stats

Top Repositories

Support Me

p2pfl's Issues

[FUTURE] GRCP

💡 The feature, motivation and pitch

communication protocol (pure TCP) to GRCP (https://grpc.io/) migration

• consider whether it would improve robustness, performance and clarity in code

Alternatives

No response

Additional context

No response

[FUTURE] Node connection during learning.

💡 The feature, motivation and pitch

Node connection during learning.

Alternatives

No response

Additional context

No response

[FUTURE] Simulation Optimization

💡 The feature, motivation and pitch

Create a new type of connections that avoids the use of TCP for local simulations.

The single instance pattern should be used and declare this operation as a new connection type.

Alternatives

No response

Additional context

No response

[FUTURE] Node authentication

💡 The feature, motivation and pitch

Node authentication to control node access to the network.

Alternatives

No response

Additional context

No response

[FUTURE] Iterative terminal for node management

💡 The feature, motivation and pitch

It would be great to manage a node through an iterative terminal.

Alternatives

No response

Additional context

No response

[FUTURE] New Federated Learning Aggregation Algorithms

💡 The feature, motivation and pitch

It would great to have a repertoire of aggregation algorithms

Alternatives

No response

Additional context

No response

[BUG] Tolerance to incomplete message reception

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

No response

Expected Behavior

Although highly improbable, it could happen that fragments of (incomplete) messages are read and produce errors.

Steps To Reproduce

No response

Anything else?

No response

In which system are you seeing the problem on?

No response

[FUTURE] Secure Agregation

💡 The feature, motivation and pitch

Include Secure Agregation for an additional privacy layer

Alternatives

No response

Additional context

No response

[FUTURE] TensorFlow support

💡 The feature, motivation and pitch

TensorFlow support

Alternatives

No response

Additional context

No response

[FUTURE] Optional Dependencies

💡 The feature, motivation and pitch

https://setuptools.pypa.io/en/latest/userguide/dependency_management.html#optional-dependencies

Alternatives

No response

Additional context

No response

Be careful with the port number

At this moment, in p2pfl/base_node.py and p2pfl/node.py, they have as default port 0. Instead, you should put something not reserve by the system and which is not on the well-known ports.

For example you can iniciate it to None and made this check before open any one.

def __init__(self, host="127.0.0.1", port=None, simulation=True):
     port = port if port is not None else choose_random_port()

[BUG] Votes not associated with rounds

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

Votes not associated with rounds

Expected Behavior

Associate votes to rounds, so that if delays occur and votes arrive when rounds have already been executed, discrepancies do not occur.

Steps To Reproduce

No response

Anything else?

No response

In which system are you seeing the problem on?

No response