Skip to content

AI-RAN and Open Source: Building Scalable Dataset Pipelines for AI Training#

Open Networking & Edge Summit 2025 digest Program Wildcard pick 31 March 2025

Abstract Recording

Alex Jinsung Choi, SoftBank Corp. (Chair, AI-RAN Alliance)

Play the recording: AI-RAN and Open Source: Building Scalable Dataset Pipelines for AI Training

  • Why: Not about the PM's domains — it's telecom RAN — but the core problem is one they know intimately from ML systems: high-quality training data is the bottleneck, and open, collaborative dataset pipelines (aggregation, curation, annotation, synthetic data, with security/compliance guardrails) are an emerging pattern for any domain where proprietary data can't be shared. Watching how the AI-RAN Alliance solves data sharing under governance constraints is a preview of how other infrastructure-ML domains will.
  • What: How open-source tooling can build collaborative dataset pipelines for AI-driven network optimization in RAN, including AI-RAN's role in synthetic data generation and its connection to global AI governance (from abstract only).
  • How: Open-source pipelines for data aggregation, curation, and annotation with security/compliance built in, plus synthetic data generation to compensate for scarce labeled telecom data.
  • Where applicable: Telecom RAN domain (SoftBank/AI-RAN Alliance-specific datasets), but the dataset-governance pattern is general.