Harmonizing Generalization and Personalization in Federated Prompt Learning
Authors: Tianyu Cui, Hongxia Li, Jingya Wang, Ye Shi
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on various datasets to explore base-to-novel generalization in both category-level and domain-level scenarios with heterogeneous data, showing the superiority of Fed PGP in balancing generalization and personalization. |
| Researcher Affiliation | Academia | 1Shanghai Tech University. Correspondence to: Ye Shi <shiye@shanghaitech.edu.cn>. |
| Pseudocode | Yes | Algorithm 1 Fed PGP |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its source code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | We selected five datasets to investigate base-to-novel class generalization ability: Oxford Pets (Parkhi et al., 2012), Flowers102 (Nilsback & Zisserman, 2008), DTD (Cimpoi et al., 2014), Caltech101 (Fei Fei, 2004), Food101 (Bossard et al., 2014). For evaluation of personalization, beyond the datasets used in base-to-novel class generalization, we employed two additional benchmark datasets: CIFAR-10 (Krizhevsky et al., 2010) and CIFAR-100 (Krizhevsky et al., 2009). |
| Dataset Splits | Yes | We equally split the datasets into base and novel classes and utilized the pathological setting by assigning a specific number of nonoverlapping base classes to each client. ... we utilize the leave-one-domain-out validation strategy. Each client model is trained on their local classes and evaluated on both local classes, base classes (classes seen on other clients), and novel classes (unseen in the whole training process). |
| Hardware Specification | Yes | All experiments are conducted with Pytorch on NVIDIA A40 GPUs. |
| Software Dependencies | No | The paper mentions 'Pytorch' but does not specify its version number or any other software dependencies with their versions. |
| Experiment Setup | Yes | Implementation Details. All methods presented in this paper are based on a frozen CLIP using two backbones, Res Net50 (He et al., 2016) and Vi T-B16 (Dosovitskiy et al., 2020), defaulting to Vi T-B16 if not explicitly specified. In federated learning, we set the client s local training epoch E = 1 and communication round T = 150 with N = 100 clients and partition rate r = 10% for CIFAR-10/CIFAR100 datasets. Besides, we consider training epoch E = 2 and communication round T = 25 with client numbers N = 10 and a full partition rate, i.e., r = 100% for other datasets. The low-rank decomposition bottleneck is set to b = 8, and the hyperparameter µ for the contrastive loss is set to 1. We employ cosine similarity as the metric function in contrastive loss. For the setting of learnable prompts, the length of prompt vectors p is 16 with a dimension of 512, token position is end with random initialization. Apart from the few-shot learning, batch sizes are set to 32 during training and 100 during testing. Additional implementation details can be found in the Appendix Section A.2. We employ SGD optimizer with learning rate η = 0.001. |