Antarctica GNSS Dataset
A comprehensive open GNSS dataset collected during a ship expedition from Europe to Antarctica, covering multiple satellite constellations and a wide range of latitudes. The dataset enables research on GNSS interference, jamming, and signal-quality degradation in polar regions, and is accompanied by the analysis code used in the Scientific Data paper.
Backreference:
- B. Bertalanič, F. Dimc, M. Bažec, A. Blatnik, “Voyage to the Frozen Continent: A Comprehensive GNSS Dataset From a Ship’s Expedition to Antarctica”, Scientific Data, 2026, (DOI)
LLM Debate Dynamics
An experimental harness for studying how teams of large language models fail and succeed when scaled. The repository hosts the configurations behind our work on the cost of consensus in homogeneous debate and the Ringelmann scaling law for effective team size, with reproducible runs on math and reasoning benchmarks across open-weight model families.
Backreference:
- B. Bertalanič, C. Fortuna, “The Cost of Consensus: Isolated Self-Correction Prevails Over Unguided Homogeneous Multi-Agent Debate”, ACM Conf. on AI and Agentic Systems, 2026, (arXiv:2605.00914)
- B. Bertalanič, C. Fortuna, “The Ringelmann Effect in Multi-Agent LLM Systems: A Scaling Law for Effective Team Size”, forthcoming
MRM3 — Machine-Readable ML Model Metadata
MRM3 is a proposed standard and reference implementation for machine-readable ML model metadata. It captures training data, evaluation conditions, energy and carbon cost, and intended deployment environment in a structured, queryable form, so that models can be discovered, audited and re-used across teams and organizations.
Backreference:
- A. Čop, B. Bertalanič, M. Grobelnik, C. Fortuna, “MRM3: Machine Readable ML Model Metadata”, MobiSys ‘25, 2025, (ACM DL, arXiv:2505.13343)
JaGuard — GNSS Jamming Position Error Correction
JaGuard corrects GNSS position errors caused by jamming using deep temporal graph networks. The repository provides training scripts, evaluation on real jamming scenarios collected with the LOG-a-TEC and Antarctica campaigns, and pretrained models ready for integration in resilient navigation pipelines.
Backreference:
- I. Kesić, A. Blatnik, C. Fortuna, B. Bertalanič, “JaGuard: Position Error Correction of GNSS Jamming With Deep Temporal Graphs”, 2025, (arXiv:2509.14000)
NAOMI — Network AI Workflow Democratization
NAOMI is an open framework for democratizing AI workflows at the network edge. It assembles data ingestion, training, validation, deployment and monitoring stages into reproducible pipelines that small and medium-size operators can run on commodity infrastructure, without depending on hyperscaler-only services.
Backreference:
Smart-Community Energy Management With LP + DRL
A hybrid optimization library that combines linear programming with multi-agent deep reinforcement learning to minimize the cost of operating energy communities. The repository provides the full training pipeline, environment definitions for shared storage and V2G assets, and evaluation against benchmark price signals.
Backreference:
- M. Pokorn, A. Čampa, M. Smolnikar, M. Mohorčič, J. Hribar, “Cost Minimization in Energy Communities With Multi-Agent Deep Reinforcement Learning and Linear Programming”, IEEE Access, 2026, (DOI)
- J. Hribar, M. Mohorčič, A. Čampa, “Improving Energy Autonomy of Positive Energy Districts Using Multi-Agent Deep Reinforcement Learning”, Scientific Reports, 2025, (DOI)
- J. Hribar, M. Mohorčič, A. Čampa, “Smart Charging of V2G-Enabled EVs in Positive Energy Districts With Shared Energy Storage”, SpliTech ‘25, (DOI)
REM-Estimate — Radio Environment Maps From Learned Elevation
REM-Estimate produces radio environment maps without LiDAR, using learned elevation models as a lightweight surrogate. The pipeline takes coarse elevation rasters and produces dense RSSI predictions that match much heavier LiDAR-based estimators, making radio mapping feasible for low-cost outdoor deployments.
Backreference:
- L. Milosheski, F. Močnik, M. Mohorčič, C. Fortuna, “Learned Elevation Models as a Lightweight Alternative to LiDAR for Radio Environment Map Estimation”, 2026, (arXiv:2604.05520)
ECAL — Carbon-Aware AI Lifecycle Simulator
eCAL is an open simulator for the energy and Carbon footprint of the AI Lifecycle in communication networks. It models data acquisition, training, deployment and inference stages, exposes their energy and carbon costs as first-class metrics, and supports what-if analysis for sustainability-driven design of AI services in 6G and beyond.
Backreference:
- S.-K. Chou, J. Hribar, V. Hanžel, M. Mohorčič, C. Fortuna, “The Energy Cost of Artificial Intelligence Lifecycle in Communication Networks”, IEEE JSAC, 2026, (DOI, arXiv:2408.00540)
- S.-K. Chou, et al., “Energy Cost of the AI/ML Workflow in O-RAN”, IEEE NFV-SDN'25
- S.-K. Chou, et al., “Towards the Standardization of Energy Efficiency Metrics of the AI Lifecycle in 6G and Beyond”, IEEE CSCN'24, (DOI)
AutoPL — Symbolic-Regression Pathloss Modeling
AutoPL is a pipeline for automatically discovering compact, interpretable pathloss models from measurement data using symbolic regression. Rather than fitting a fixed empirical formula, it searches over closed-form expressions that balance accuracy, complexity and physical plausibility, yielding models that propagation engineers can read and reason about.
Backreference:
- A. Anaqreh, S.-K. Chou, B. Bertalanič, M. Mohorčič, T. Lagkas, C. Fortuna, “Automated Modeling Method for Pathloss Model Discovery”, 2026, (IEEE Trans. Mobile Comput., arXiv:2505.23383)
- A. Anaqreh, et al., “Towards Automated and Interpretable Pathloss Approximation Methods”, AAAI ‘25 Workshop on AI for Wireless Communications and Networking, 2025
ZeroUED — Zero-Shot Self-Supervised Unknown Emitter Detection
ZeroUED is a self-supervised pipeline for detecting previously unseen radio emitters from raw I/Q recordings, without labeled training data. The release includes the encoder, the design-principle ablations from the paper and the configurations used to identify novel devices in the wild.
Backreference:
- M. Krasnov, L. Milosheski, M. Mohorčič, C. Fortuna, “Design Principles of Zero-Shot Self-Supervised Unknown Emitter Detectors”, 2025, (arXiv:2511.07026)
- M. Krasnov, et al., “Novel Devices Identification With Deep Clustering”, IEEE ICMLCN'25, (IEEE Xplore)
- L. Milosheski, B. Bertalanič, C. Fortuna, M. Mohorčič, “Radio Signals Recognition With Unsupervised Deep Learning: A Survey”, IEEE Access, 2025, (DOI)
Indoor Radio Mapping Dataset (3D Point Cloud + RSSI)
An open indoor measurement dataset that fuses dense 3D point-cloud geometry with co-located RSSI measurements, enabling multimodal research on radio environment maps, geometry-aware localization and reproducible benchmarking across indoor environments.
Backreference:
- L. Milosheski, K. Akiyama, B. Bertalanič, J. Hribar, R. Shinkuma, “An Indoor Radio Mapping Dataset Combining 3D Point Clouds and RSSI”, 2025, (arXiv:2511.00494)
KAN-TS: Kolmogorov–Arnold Networks for Time-Series Classification
Reference implementation and evaluation harness for using Kolmogorov–Arnold Networks (KANs) as an inherently interpretable alternative to MLP heads in time-series classification pipelines. The repository ships training scripts, ablations and the UCR-benchmark configurations used in the paper.
Backreference:
TS-Graph-Segmentation
A reproducibility package for granular time-series segmentation via a network-science lens. Time series are projected onto visibility graphs and segmented using Louvain community detection, producing change points and segment-level features that are competitive with deep baselines while remaining transparent and parameter-light.
Backreference:
- I. Kesić, C. Fortuna, M. Mohorčič, B. Bertalanič, “A Network Science Approach to Granular Time Series Segmentation”, 2025, (arXiv:2505.17640)
- I. Kesić, et al., “Visibility Graphs-Based Time Series Segmentation With Louvain Community Detection”, ERK'24
Electricity Knowledge Graph Datasets
The repository includes downloads for the datasets and all the neccesary code to run the pipeline for preprocessing the data and generating the knowledge graph. The knowledge graph is generated from a set of raw datasets containing electricity consumption data from multiple regions and households. The data is preprocessed and harmonized to generate a knowledge graph containing information about the households, appliances, and electricity consumption. We also provide a model training pipeline that can be used to train a model for on/off appliance classification.
Smart Home Energy Trading With Deep Reinforcement Learning
This python library storest code for a conference paper about smart home energy management using Deep Reinforcement Learning. It is possible for the user to train their own DRL agent, as well as load pre-trained models and test the agent’s performance.
Backreference:
- M. Pokorn, et al., “Smart Home Energy Cost Minimisation Using Energy Trading with Deep Reinforcement Learning”, 2023, (ACM BuildSys'23)
CCWEBAPP
This tool helps estimate the computational complexity of neural networks. It implements the computations according to the methodology explained in the IEEE ICC paper linked below. It supports fully connected, convolutional, and pooling layers.
Backreference:
- A. Pirnat, et al., “Towards Sustainable Deep Learning for Wireless Fingerprinting Localization”, 2022, (IEEE ICC'22), (arXiv:2201.09071)
BLE Fingerprints Dataset
The available dataset contains received signal strength (RSS) measurements made with Bluetooth Low Energy (BLE) technology, which can be used for outdoor fingerprinting based localization applications. The dataset was collected with 25 nodes of the LOG-a-TEC testbed positioned at the campus of the Jozef Stefan Institute, Ljubljana.
Backreference:
- B. Bertalanič, et al., “LOG-a-TEC Testbed Outdoor Localization using BLE Beacons”, 2022, (BalkanCom'22)
Data-Driven Link Quality Estimation
This repository hosts the source code used in a comprehensive tutorial and survey paper focused on link quality estimation research. The materials contained herein provide a detailed overview and practical applications of various methodologies and technologies employed in the study of link quality estimation. This repository serves as a valuable resource for researchers and practitioners in the field, offering insights and tools developed as part of this significant research effort.
Backreference:
- G. Cerar, et al., “Machine Learning for Wireless Link Quality Estimation: A Survey”, 2021, (IEEE COMST)
- M. Kulin, et al., “Data-Driven Design of Intelligent Wireless Networks: An Overview and Tutorial”, 2016, (MDPI Sensors)
Kubitect
Kubitect is an open source project that aims to simplify the deployment and subsequent management of Kubernetes clusters. It provides a CLI tool written in Golang that lets you set up, upgrade, scale, and destroy Kubernetes clusters. Under the hood, it uses Terraform along with terraform-libvirt-provider to deploy virtual machines on target hosts running libvirt. Kubernetes is configured on the deployed virtual machines using Kubespray, the popular open source project.
Backreference:
- C. Fortuna, et al., “On-Premise Artificial Intelligence as a Service for Small and Medium Size Setups”, 2024, (arXiv:2210.06956)
UWB Localization Dataset
UWB localization data set contains measurements from four different indoor environments. The data set contains measurements that can be used for range-based localization evaluation in different indoor environments using 9 DW1000 UWB transceivers (DWM1000 modules) connected to the networked RaspberryPi computer using in-house radio board SNPN_UWB. 8 nodes were used as localization anchor nodes with fixed locations in individual indoor environment and one node was used as a mobile localization tag.
Backreference:
- K. Bregar, M. Mohorcic, “Improving Indoor Localization Using Convolutional Neural Networks on Computationally Restricted Devices”, 2018, (IEEE Access)
LPWAN Trace Set
A dataset containing 24 hours of continuous spectrum measurements. A proprietary spectrum sensing device placed on top of a building in a mid-sized European city recorded 5 PSD measurements per second using 1024 FFT bins in a 192 kHz wide band inside the unlicensed European 868 MHz SRD band.
Backreference:
- C. Fortuna, et al., “Automatic Detection and Query of Wireless Spectrum Events from Streaming Data”, 2018, (arXiv:1804.05019)
- T. Gale, et al., “Automatic Detection of Wireless Transmissions”, 2022, (IEEE Access)
LOG-a-TEC
LOG-a-TEC is a diverse testbed used for research purposes. It started in 2016 and evolved overtime into its third iteration. It covers ultra narrow band and ultra wide band, packet based experimentation, clean slate protocol design, composable and modular protocol stacks, custom and advanced spectrum sensing and signal generating functions in sub-GHz spectrum.
