Proof of Concepts

PoC 1: Data management and experiment on demand

The aim of this PoC is to showcase the capabilities of the 6G-DALI AI framework to satisfy
users’ requests for 6G data by building the concept of the 6G Data Space, with the participation
of all 3 6G testbeds and the TID cloud. The two pipelines supported by 6G-DALI for dataset
search will be demonstrated. The first one aims to demonstrate the case that the data is already in
the 6G Data Space and discovered through the AI framework using the concept of Gaia-X
service catalogue, while the second one showcases the experiment(s) triggering the case when
the data is not existing in the 6G Data Space. Both experiments will demonstrate the ELT
pipeline, from extracting the semantics of users to translating it to a service catalogue request,
and if the dataset is not available translate the request to experiments to run on one or multiple
testbeds connected to the project. 
The PoC’s objectives are: 

  • Demonstrate the integration of the AI framework with 6G Data Space and Gaia-X service
    catalogue, authorization, and trust. 
  • Demonstrate the AI framework’s ability to translate user requests to data sets to experiments
    via LLM technologies. 
  • Demonstrate the ELT pipeline to populate, transform (data cleaning and augmentation), and
    load to the data space for data analytic collection and storage.

PoC 2: AIaaS for CDN apps via cross-testbed decentralized MLOps

This PoC aims at validating the functionalities of the 6G-DALI e2e AI framework with
decentralized MLOps, meta-orchestration and AI experimentation services, specifically targeting
the assessment of its cooperative and distributed capabilities for the management of ML models
and showcasing its applicability at a CDN vertical application. In particular, the PoC will
validate the 6G-DALI unified approach for managing and executing ML tasks and processes
with full automation, streamlining the various steps and taking autonomous decisions related to
AI/ML models training, deployment, placement, hyperparameter optimization, and transfer
learning/quantization as a service. Moreover, the PoC aims at assessing the capability of the 6G-
DALI e2e AI framework to rely on heterogeneous testbed infrastructures with different MLOps
software stacks, demonstrating a solution that is independent from the underlying computing
technologies (meta-Orchestration concept) and from the user and ML model constraints. 
The objectives of of this PoC can be summarized as follows: 

  • Demonstrate full automation in managing and performing ML processes and workflows,
    specifically validating streamlined training, deployment, inference, testing and validation tasks. 
  • Demonstrate the capability to execute AI/ML tasks and processes in a transparent and unified
    way on top of different testbed infrastructure and MLOps stacks.
  • Demonstrate the capability to deploy and place AI/ML tasks and workloads across the extreme-
    edge, edge, cloud continuum, according to user, data and AI/ML requirements via model
    compression and quantization. 
  • Demonstrate the capability to support distributed AI/ML tasks and workloads, such as those
    required in case of federated learning, cooperative AI/ML inference, and transfer learning. 
  • Demonstrate the capability to automatically detect model drifts and perform corrective actions
    according to user, data and AI/ML requirements.
PoC 3: DTT and RLOPS for large and medium-scale experiments

This PoC showcases 6G-DALI capabilities to run large-scale experiments using the AI
framework and the DT, with two-fold objectives. First, datasets are generated and integrated into
the 6G dataspace, using 6G-DALI ETL. Second, realize RLOps to test and validate trained RL
agents. 
The objectives of this PoC are: 

  • Demonstrate the automatic deployment of large-scale experimentation on the 6G-DALI from
    the definition, deployment, and generation of datasets involving the AI framework components:
    user gateway, data acquisition, data cleaning, adaptation layer, and DTT adapter. 
  • Demonstrate the configuration and data set collection using the OEM DTT and xApp. 
  • Demonstrate the DT capabilities to generate Medium and large-scale experimentations with
    low and high-fidelity DT. 
  • Demonstrate the integration of the DTT via ELT to the 6G Data Space. 
  • Demonstrate RLOps mechanisms devised in 6G-DALI.

Scroll to Top