Exploiting Shared Representations for Personalized Federated Learning
Authors: Liam Collins, Hamed Hassani, Aryan Mokhtari, Sanjay Shakkottai
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide extensive experimental results demonstrating the improvement of our method over alternative personalized federated learning approaches in heterogeneous settings. Through a combination of synthetic and real datasets (CIFAR10, CIFAR100, FEMNIST, Sent140) we show the benefits of Fed Rep |
| Researcher Affiliation | Academia | 1University of Texas at Austin 2University of Pennsylvania. Correspondence to: Liam Collins <liamc@utexas.edu>. |
| Pseudocode | Yes | Algorithm 1 Fed Rep |
| Open Source Code | No | The paper does not include a statement about releasing open-source code or a link to a code repository. |
| Open Datasets | Yes | We use four real datasets: CIFAR10 and CIFAR100 (Krizhevsky et al., 2009), FEMNIST (Caldas et al., 2018; Cohen et al., 2017) and Sent140 (Caldas et al., 2018). |
| Dataset Splits | No | The paper describes training on various datasets and uses a test set for evaluation, but does not explicitly mention a separate validation split or how one was handled for hyperparameter tuning. |
| Hardware Specification | No | The paper does not specify the hardware (e.g., GPU, CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions software like CNNs, MLP, RNN as models, but does not specify programming languages or library versions (e.g., Python, PyTorch, TensorFlow versions) used for implementation. |
| Experiment Setup | Yes | Parameters: Participation rate r, step sizes α, η; number of local updates τ; number of communication rounds T. Initialize φ0, h0 1, . . . , h0 n... In each case, Fed Rep executes ten local epochs of SGD with momentum to train the local head, followed by one or five epochs for the representation, in each local update (depending on the dataset). All other methods use the same number of local epochs as Fed Rep does for updating the representation. |