Federated Visual Classification with Real-World Data Distribution (2003.08082v3)

Abstract: Federated Learning enables visual models to be trained on-device, bringing advantages for user privacy (data need never leave the device), but challenges in terms of data diversity and quality. Whilst typical models in the datacenter are trained using data that are independent and identically distributed (IID), data at source are typically far from IID. Furthermore, differing quantities of data are typically available at each device (imbalance). In this work, we characterize the effect these real-world data distributions have on distributed learning, using as a benchmark the standard Federated Averaging (FedAvg) algorithm. To do so, we introduce two new large-scale datasets for species and landmark classification, with realistic per-user data splits that simulate real-world edge learning scenarios. We also develop two new algorithms (FedVC, FedIR) that intelligently resample and reweight over the client pool, bringing large improvements in accuracy and stability in training. The datasets are made available online.

Author Guidelines for ECCV Submission: An Analytical Overview

This document provides a detailed guideline for authors preparing submissions for the European Conference on Computer Vision (ECCV). It outlines various critical aspects of the submission process, including formatting guidelines, anonymity requirements, and policy adherence. This essay summarizes its key components and reflects on its implications for the academic community, particularly in terms of maintaining the rigor and integrity in scholarly publishing.

Initial Submission Specifications

The document underscores the importance of adhering to English language standards and emphasizes strict compliance with paper length requirements. Submissions must not exceed 14 pages, excluding references. This restriction ensures uniformity and prevents authors from manipulating section lengths to circumvent the page limit. Noncompliance with these formatting rules results in immediate rejection. This stringent approach underscores ECCV's commitment to maintaining high evaluative standards.

Anonymity and Confidentiality

The paper delineates the double-blind review process, explaining that both authors and reviewers remain anonymous to each other. The authors must avoid self-identification in citations, refraining from using first-person pronouns. The procedures for an anonymous submission emphasize objectivity, reducing reviewer bias based on author identification. This practice is crucial for ethical academic review processes and fosters a merit-based evaluation of submitted research.

Policies on Dual Submission and Blind Review

The manuscript provides clear guidelines regarding dual and double submissions. Authors are prohibited from submitting substantially similar content to multiple venues simultaneously, ensuring that ECCV disseminates original contributions. While authors must cite concurrent submissions, they must also provide different contexts and results to underscore each work's novelty. These rules aim to preserve academic novelty and prevent redundant reviewer workload, enhancing the overall integrity and efficiency of the review process.

Formatting and Manuscript Preparation

Detailed instructions are given on formatting and preparing the manuscript, with a strong recommendation for using LaTeX. The guidelines cover aspects such as layout, heading styles, and the inclusion of figures and equations. The emphasis on consistent formatting ensures that all submissions meet a professional standard, facilitating reader comprehension and comparative evaluation during review.

Practical and Theoretical Implications

The document, although procedural, implicitly contributes to the theoretical discourse on scientific communication. By mandating rigorous adherence to submission guidelines, ECCV supports a foundational framework for scholarly exchange. Practically, these guidelines help streamline the review process for a conference that draws a large volume of submissions, ensuring that reviewers spend their time evaluating the scientific content, not deciphering inconsistent formatting.

Future Directions in AI Conference Submissions

The document's comprehensive structure anticipates trends toward even more automated and technically demanding submission systems. With artificial intelligence evolving, future conferences might integrate AI-driven tools for checks on formatting compliance, plagiarism, and even preliminary content evaluation to streamline the initial submission stage further.

This guideline document for ECCV captures an essential blueprint for structured academic submissions, illustrating the importance of uniformity, transparency, and ethical standards in the research dissemination process. By setting these high standards, ECCV continues to contribute significantly to the field of computer vision and the broader scientific community.