Federated Learning from Vision-Language Foundation Models: Theoretical Analysis and Method

Authors: Bikang Pan, Wei Huang, Ye Shi

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental These theoretical claims have been further supported by empirical experiments. Our code is available at: https://github.com/Pan Bikang/Prompt Folio.git. ... In this section, we conduct experiments with the CLIP model to empirically demonstrate the performance advantages of Prompt Folio.
Researcher Affiliation Academia Bikang Pan Shanghai Tech University panbk2023@shanghaitech.edu.cn Wei Huang RIKEN Center for Advanced Intelligence Project wei.huang.vr@riken.jp Ye Shi Shanghai Tech University shiye@shanghaitech.edu.cn
Pseudocode Yes Algorithm 1 (Prompt Folio) Global-Local Prompt Portfolio
Open Source Code Yes Our code is available at: https://github.com/Pan Bikang/Prompt Folio.git.
Open Datasets Yes The experiment is conducted on the CIFAR-100 dataset by default, ... We use CIFAR-100 [25], Domain Net [36], Office-Caltech10 [15], Oxford Pets [35], and DTD [11]
Dataset Splits No The paper mentions using standard datasets like CIFAR-100 but does not explicitly state the train/validation/test splits, their percentages, or sample counts, nor does it explicitly reference predefined splits.
Hardware Specification Yes The experiments are conducted on a cluster with 2 Intel Xeon 5218R, 512GB memory, and 8 NVIDIA Tesla A40 GPUs 48GB.
Software Dependencies No The paper does not provide specific version numbers for key software components such as programming languages, libraries, or frameworks used in the experiments.
Experiment Setup Yes The experiment is conducted on the CIFAR-100 dataset by default, with the model trained for 10 epochs locally and the results evaluated over 100 communication rounds.