Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Budgeted Online Model Selection and Fine-Tuning via Federated Learning

Pouya M. Ghari, Yanning Shen

TL;DR

This work tackles online model selection with a large dictionary of models on a server while edge clients face memory constraints. It introduces OFMS-FT, a federated framework where clients maintain a subset of models, perform online selection via a multiplicative-weight mechanism, and jointly fine-tune models under bandwidth limits. Theoretical regret bounds show sub-linear performance for both clients and the server, with tighter bounds as client budgets and communication bandwidth increase. Empirical results on image classification and regression tasks demonstrate that OFMS-FT outperforms several baselines, validating the practical impact of memory-aware, federated model selection and fine-tuning in non-stationary, heterogeneous data settings.

Abstract

Online model selection involves selecting a model from a set of candidate models 'on the fly' to perform prediction on a stream of data. The choice of candidate models henceforth has a crucial impact on the performance. Although employing a larger set of candidate models naturally leads to more flexibility in model selection, this may be infeasible in cases where prediction tasks are performed on edge devices with limited memory. Faced with this challenge, the present paper proposes an online federated model selection framework where a group of learners (clients) interacts with a server with sufficient memory such that the server stores all candidate models. However, each client only chooses to store a subset of models that can be fit into its memory and performs its own prediction task using one of the stored models. Furthermore, employing the proposed algorithm, clients and the server collaborate to fine-tune models to adapt them to a non-stationary environment. Theoretical analysis proves that the proposed algorithm enjoys sub-linear regret with respect to the best model in hindsight. Experiments on real datasets demonstrate the effectiveness of the proposed algorithm.

Budgeted Online Model Selection and Fine-Tuning via Federated Learning

TL;DR

This work tackles online model selection with a large dictionary of models on a server while edge clients face memory constraints. It introduces OFMS-FT, a federated framework where clients maintain a subset of models, perform online selection via a multiplicative-weight mechanism, and jointly fine-tune models under bandwidth limits. Theoretical regret bounds show sub-linear performance for both clients and the server, with tighter bounds as client budgets and communication bandwidth increase. Empirical results on image classification and regression tasks demonstrate that OFMS-FT outperforms several baselines, validating the practical impact of memory-aware, federated model selection and fine-tuning in non-stationary, heterogeneous data settings.

Abstract

Online model selection involves selecting a model from a set of candidate models 'on the fly' to perform prediction on a stream of data. The choice of candidate models henceforth has a crucial impact on the performance. Although employing a larger set of candidate models naturally leads to more flexibility in model selection, this may be infeasible in cases where prediction tasks are performed on edge devices with limited memory. Faced with this challenge, the present paper proposes an online federated model selection framework where a group of learners (clients) interacts with a server with sufficient memory such that the server stores all candidate models. However, each client only chooses to store a subset of models that can be fit into its memory and performs its own prediction task using one of the stored models. Furthermore, employing the proposed algorithm, clients and the server collaborate to fine-tune models to adapt them to a non-stationary environment. Theoretical analysis proves that the proposed algorithm enjoys sub-linear regret with respect to the best model in hindsight. Experiments on real datasets demonstrate the effectiveness of the proposed algorithm.
Paper Structure (15 sections, 3 theorems, 67 equations, 1 figure, 5 tables, 2 algorithms)

This paper contains 15 sections, 3 theorems, 67 equations, 1 figure, 5 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Online Federated Model Selection
  4. Online Federated Model Selection
  5. Online Model Fine-Tuning
  6. Regret Analysis
  7. Related Works and Discussions
  8. Experiments
  9. Conclusion
  10. First Fit Decreasing Algorithm
  11. Proof of Theorem \ref{['th:4']}
  12. Supplementary Experimental Results and Details
  13. Supplementary Discussions and Analysis
  14. Supplementary Analysis
  15. Proof of Theorem \ref{['th:5']}

Key Result

Theorem 1

Assume that clients utilize the first fit decreasing algorithm in order to split models into clusters ${\mathbb{D}}_{i1,t}, \ldots, {\mathbb{D}}_{im_{i,t},t}$. Under (a1) and (a2), the expected cumulative regret of the $i$-th client using OFMS-FT is bounded by which holds for all $i \in [N]$. Under (a1)--(a4), the cumulative regret of the server in fine-tuning model $k$ using OFMS-FT is bounded b

Figures (1)

  • Figure 1: Average regret of clients using OFMS-FT with the change in budget $B_i$.

Theorems & Definitions (6)

  • Theorem 1
  • proof
  • Lemma 2
  • proof
  • Theorem 3
  • proof