lunan0320/federated-learning-knowledge-distillation

repo in several methods FedAvg, FedMD, FedProto, FedProx, FedHKD.

Hi there, I'm lunan0320 👋

Federated Learning Researcher | Privacy Tech Enthusiast

As a passionate student in the field of Cybersecurity, my research is driven by a fascination with Federated Learning, the acceleration of distributed encrypted training, and fine-tuning of large language models with a strong focus on privacy-preserving techniques such as Homomorphic Encryption and Differential Privacy.

Outside the digital realm, I channel my energy into swimming and fitness.

What I'm working on 🚀

• Improving efficiency and security in Federated Learning systems.
• Exploring advanced encryption methods for privacy in AI, including Homomorphic Encryption.
• Investigating Differential Privacy applications in fine-tuning large language models.

Tech Stack 🛠️

• Python, PyTorch, C++, Java, C
• Homomorphic Encryption Libraries
• Differential Privacy Frameworks
• CUDA & Parallel Computing Platforms

Thanks for stopping by, and please feel free to check out my repositories below!

---

• 👉 GitHub Blog

• 👉 CSDN Blog

federated-learning-knowledge-distillation's Issues

Please consult the author

I would like to ask, if I change the number of clients to 50, what other modifications are needed to ensure that the code runs

Has the code been reproduced

Dear friend,
Hello friends, have you reproduced this code, why do I reproduce the code accuracy at the highest time is only 77, and there is another problem, reproduce --dataset SVHN --batch_size 64 --num_epochs 50 --clip_grad 1.1 --lr 0.001 --num_clients 10 --num_classes 10 --sampling_rate 1 --local_ ep 5 --beta 0.5 --seed 0 --code_len 32 --alg FedHKD --part 0.1 --model shufflenet --temp 0.5 This parameter, why does loss2:Inf always appear? The -lr_sh_rate parameter he gave doesn't seem to be used. If you reproduce it, can you tell me the parameters you use, and this loss2:inf problem, if you know how to solve it, can you tell me?
Looking forward to hearing from you.