Machine Learning in Telecom

𝐀𝐬 𝐰𝐞 𝐬𝐭𝐞𝐩 𝐢𝐧𝐭𝐨 𝟐𝟎𝟐𝟔, 𝐈 𝐛𝐞𝐥𝐢𝐞𝐯𝐞 𝐭𝐡𝐢𝐬 𝐰𝐢𝐥𝐥 𝐛𝐞 𝐭𝐡𝐞 𝐲𝐞𝐚𝐫 𝐰𝐡𝐞𝐫𝐞 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 𝐭𝐫𝐮𝐥𝐲 𝐜𝐨𝐦𝐞𝐬 𝐨𝐟 𝐚𝐠𝐞 𝐢𝐧 𝐭𝐞𝐥𝐞𝐜𝐨𝐦! From proof-of-concept to production-ready implementation, the industry is poised for a shift toward Autonomous Network intelligence 🌐🔁 In my experience working with telecom leaders, the complexity of 5G standalone networks has exposed the limitations of conventional automation. The numbers are compelling: Agentic AI can deliver an 80% reduction in time spent on analysis and decision-making, while early adopters report 20-30% improvements in network efficiency. 𝐖𝐡𝐲 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 𝐌𝐚𝐭𝐭𝐞𝐫𝐬 𝐍𝐨𝐰 TM Forum's autonomous network framework defines Level 5 as complete end-to-end autonomy without human intervention - and Agentic AI is the critical enabler to reach this milestone. 𝐊𝐞𝐲 𝐂𝐡𝐚𝐫𝐚𝐜𝐭𝐞𝐫𝐢𝐬𝐭𝐢𝐜𝐬 ▶️ Specialized AI Agents: Multiple agents analyze over 60,000 KPIs and identify up to 20 distinct classes of network issues with minimal false positives. ▶️ 'Talk to Network' Interface: Natural language interactions transform operations. Operators ask "What's going on?" or "What needs fixing?" and receive coherent, actionable responses. ▶️ Supervisor Agent Orchestration: A GenAI-powered coordinator manages the ecosystem, aligning actions with business objectives. ▶️ Closed-Loop Automation: The system moves from reactive troubleshooting to proactive optimization - often without human intervention. ▶️ Cloud Integration: Leveraging AWS infrastructure including Amazon Bedrock for high-performance computing and enterprise-grade security. 𝐈𝐧𝐝𝐢𝐚𝐧 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 What excites me most is seeing this shift already underway in India. Almost all Indian telecom operators are evaluating or deploying AI solutions: ▶️ Jio has established a dedicated AI subsidiary and is building gigawatt-scale data centers to power its "Jio Brain" platform. ▶️ airtel deploys AI agents for billing and customer care, with predictive maintenance improving network efficiency by up to 30%. ▶️ Vodafone Idea Limited has rolled out AI-powered spam detection flagging over 600 million spam calls. 𝐓𝐡𝐞 𝐏𝐚𝐭𝐡 𝐅𝐨𝐫𝐰𝐚𝐫𝐝 For telco service providers, adopting Agentic AI means efficiently managing 5G complexity, containing costs, and delivering superior customer experiences. This also represents significant opportunity for Indian technology firms, GCCs, and consulting organizations. The journey toward Level 5 Autonomous Networks has begun. I'm convinced that those who embrace this evolution now will set the standard for the industry. Curious to hear your perspectives on this evolution 💡 #CapgeminiInvent #Telecom #AI #Leadership #DigitalTransformation #MakeItReal Praveen Shankar | Frederic V. | Sandeep Arora | Alessandro Puglia | Abhishek Soni | Sarvesh Bhatnagar

Some moments in life remind you that the journey is just as important as the destination. Grateful for the people who make learning, growth, and hard work enjoyable. Behind every project is the support, laughter, and encouragement that keep us going. Telecom companies spend millions acquiring new customers, yet many leave within their first six months. What if businesses could predict who is likely to leave and take action before they do? That’s exactly what I set out to solve using Machine Learning and #datascience. I built a customer retention prediction model using real-world telecom data to uncover patterns behind service cancellations and help businesses retain more customers. I started with exploratory data analysis to identify key trends influencing customer drop-off. Feature engineering played a huge role in transforming tenure, contract types, payment methods, and internet usage into meaningful insights. To improve prediction accuracy, I balanced the dataset using #SMOTE and tested multiple machine learning models, including Logistic Regression, KNN, Decision Trees, and Random Forest. After rigorous testing, Random Forest with fully engineered features delivered the best performance, achieving a ROC-AUC score of 0.845. It effectively identified at-risk customers with a recall of 79.2 percent while maintaining a precision of 51.8 percent to reduce false alarms. This project is not just about building models but about making Machine Learning work for real business problems. Turning raw data into actionable insights is where the real impact happens. GitHub Repository: [Customer Retention Prediction](https://lnkd.in/d4e6p_M4) #MachineLearning #DataScience #CustomerRetention #PredictiveAnalytics #Python #AI #FeatureEngineering #RandomForest #BusinessStrategy

Every industry is being reshaped by #AI. But #telecoms? Many still write it off as too slow, too legacy, too regulated. That view is increasingly outdated. Following #MWC2026, I mapped Ericsson's AI deployments across the standard AI stack: Infrastructure, Model, Platform and Application. The picture is more complete than most people realize. This is not a roadmap. Most of it is live today. Here is what stands out: 📡 Infrastructure (Device & RAN) AI embedded at the physical layer. On-device inference, Vehicle-to-Everything (V2X) in connected cars, analytics inside IoT sensors. In the RAN, spectrum optimized continuously, energy cut dynamically, beamforming improved in real time. The network hardware is becoming intelligent. 🧠 Model (Multi-access Edge Computing (MEC)) Where AI models actually run, close to the source with single-digit millisecond latency. Autonomous fault detection, real-time inference, industrial automation, live network simulation. From reactive operations to self-healing behavior. 🏛 Platform / Tooling (5G Core) Orchestration, slicing, policy and APIs all AI-driven. Operators declare intent. AI configures the rest. The role of the network engineer shifts from manual configuration to oversight. ☁️ Application (Cloud & Operations Support Systems (OSS)) AI running operations end-to-end. Predicting failures, automating planning, moving humans to oversight. Federated learning and an AI model marketplace are next. Ericsson is not adding AI to the network. They are rebuilding the network around it. 🔭 Looking ahead Most AI transformations sit on top of infrastructure. In telecoms, it is happening inside it. Near term, cross-layer AI and open Network APIs turn the telecoms stack into a platform others build on. By 2030, 6G makes AI-nativeness a design requirement, not a retrofit. The network stops carrying intelligence and becomes the intelligence layer itself. Telecoms is not catching up to the AI wave. It is becoming the infrastructure the AI wave runs on. Proud to be part of building exactly that at Ericsson. 💡 Which layer of the stack surprises you most? #Telecoms #AI #5G #MWC #Ericsson #NetworkIntelligence #AIStack #6G

“I don’t need GPU programming. Telecom doesn’t use that stuff.” A senior architect told me this years ago when I mentioned I was learning CUDA. And honestly? He wasn’t wrong… at the time. It reminded me of telecom in the 4G era — everything was boxes, hardware, fixed functions, and purpose-built silicon. Why learn parallel computing when the network ran on appliances? But here’s what fixed-function networks can’t do: → Scale AI-native workloads → Run real-time inference → Handle Massive MIMO at 6G levels → Simulate entire networks as digital twins → Accelerate UPF, beamforming, LDPC, and sensing And here’s the twist… CUDA is what makes all of that possible. Because CUDA is basically a superpower: It lets you use a GPU not just for graphics — but for running thousands of operations at the same time. Perfect for: AI workloads, DSP workloads, Packet processing, RAN PHY acceleration. 6G research and simulation. That’s why operators chasing 6G — NTT DoCoMo, SK Telecom, Vodafone, AT&T, Deutsche Telekom — are already building compute-native networks powered by GPUs. And here’s the best part… You don’t need a C++ background to start. Modern tools let telecom engineers learn GPU acceleration with Python: → CUDA Python → Triton → PyTorch → NVIDIA Aerial for RAN/Core → TensorRT for real-time inference Because CUDA isn’t just a programming model, It’s how telecom turns AI, DSP, and packet processing into accelerated, scalable, software-defined systems. 6G won’t be built by people who “configure networks.” It will be built by people who understand telecom + compute + AI. Resources to Begin 🔗 CUDA Basics https://lnkd.in/g4Rkpj22 🔗 CUDA Python (no C++ needed) https://lnkd.in/gJ8ErQX6 🔗 NVIDIA Aerial (Telecom Acceleration) https://lnkd.in/gFkMKcxF 🔗 Beginner Parallel Programming Guide https://lnkd.in/gVP_gswD 6G won’t just be built by telecom engineers — it will be built by those who understand telecom + AI + accelerated compute. #tahasajid #6g #GPU #CUDA #telecom

At MWC Barcelona this year, we launched the GSMA Open-Telco LLM Benchmarks to unite a community tackling the unique challenges of telecom AI. The first results were clear: out-of-the-box AI models simply aren’t fit for telco-specific needs. Now, with version 2.0, this effort has evolved into a thriving, open-source collaboration. The findings point to a hybrid architecture as the most effective path forward - combining the broad reasoning of foundation models with the precision of specialised components. In addition to providing clear direction for AI in telecom, what’s really exciting is the unprecedented level of industry collaboration. Operators including AT&T, China Telecom Global, Deutsche Telekom, du, KDDI Corporation, KPN, Liberty Global, Orange, Telefónica, Turkcell, Swisscom, and Vodafone are joined by research and technology partners - Adaptive AI, Datumo, Huawei GTS, Hugging Face, The Linux Foundation, Khalifa University, NetoAI, Universitat Pompeu Fabra - Barcelona (UPF), The University of Texas at Dallas and Queen's University - to build a shared ecosystem for experimentation, validation, and learning. Read more in our latest blog: https://lnkd.in/eTDH5PBX

🚦 **Reflections from NVIDIA GTC Washington, D.C 2025.** Last week’s GTC made one thing clear; AI-native infrastructure is evolving fast, and telecom is being invited to the table. But amid the excitement, it’s worth taking a balanced look at what’s real today versus what’s aspirational. 📡 Telecom in the Spotlight - **Nokia and NVIDIA** announced work on *AI-native 6G RAN nodes* using the Aerial/ARC-Pro platform, a promising signal of how compute and connectivity are converging. - Huang emphasized that *telecom is the nervous system of the economy*, calling for greater technology independence and domestic innovation. - Panels on “AI for Telecommunications” showcased prototypes of intelligent RAN optimization, edge analytics, and network planning powered by machine learning. ⚖️ Signals vs. Substance - **Early days**: Many of these initiatives are still in the *proof-of-concept* phase. Integrating AI models into live RAN environments will require years of testing, spectrum-policy clarity, and vendor alignment. - **Cost and complexity**: Embedding GPUs and AI accelerators into network nodes could shift the economics of telecom infrastructure, it’s a good idea, but not a trivial retrofit. Also, we have been there before with the whole MEC concept (which failed). - **Governance**: As sovereign-tech conversations grow louder, telcos will need to navigate new compliance, data-sovereignty, and security frameworks before large-scale deployment. 💭 My Take AI-enabled wireless is an exciting frontier, it promises smarter, more adaptive networks. .....But for now, the prudent path is **experimentation with guardrails**: pilot at the edge, validate the economics, and align architecture standards before scaling. If you’re in telecom or enterprise network architecture, this is a space to watch closely and approach "thoughtfully". #NVIDIAGTC #Telecom #AI #6G #RAN #EdgeComputing #NetworkTransformation

"Which AI model is best for telecom?" Finally, there's a real answer. GSMA just launched the Open-Telco LLM Benchmark. And it changes everything. WHAT IS IT? First AI evaluation framework built FOR telecom Not general trivia tests. Not consumer chatbot benchmarks. Real telecom tasks: → Network configuration → Log troubleshooting → 3GPP standards interpretation → Policy logic → Schema-driven operations THE 5 TEST DATASETS: 1. TeleYAML - Configuration file generation 2. TeleLogs - Network log analysis 3. TeleMath - RF calculations, capacity planning 4. TeleQnA - Telecom domain Q&A 5. 3GPP-TSG - Standards compliance & interpretation Tests what matters: Precision, reasoning, telecom context WHY THIS MATTERS: Before: → Operators testing ChatGPT, Claude, Gemini → No standardized way to compare → "Seems good" = evaluation methodology → Wasting millions on wrong models Now: → Objective scores on telecom tasks → Know which model handles YANG configs → Which understands 3GPP specs → Which can debug network logs AI hype → Engineering data THE PROBLEM IT SOLVES: General benchmarks (MMLU, ARC): → Test trivia, common knowledge → "What's the capital of France?" → Useless for telecom Can't evaluate: → Schema-driven syntax → Standards compliance → RF engineering math → Network troubleshooting logic Open-Telco Benchmark: → Tests telecom-specific intelligence → Real operational scenarios → Pass = Actually useful for networks WHAT'S NEXT: Operators can now: → Test models against same benchmark → Compare vendors objectively → Make data-driven AI decisions → Stop guessing Vendors must: → Optimize for telecom tasks → Publish benchmark scores → Compete on measurable performance THE BOTTOM LINE: GSMA just gave telecom an AI yardstick. No more: → "ChatGPT seems smart" → "Claude sounds good" → "Gemini might work" Now: → Model X scores 87% on TeleYAML → Model Y scores 92% on 3GPP-TSG → Make informed decisions AI for telecom just became measurable. Finally. Using AI in your network? → 🧪 Tested models? → 📊 Need benchmark? → 🤔 Which model winning? Share below 👇 Join my Free 5G/6G Learning Free whatsapp Channel : https://lnkd.in/gerTY-kr ♻️ Repost this to help your network get started ➕ Follow Nitin Gupta for more

Can AI agents safely manage the critical infrastructure of future 6G networks? Happy to share our paper that tackles a hidden but dangerous flaw in how AI makes decisions, especially when it comes to mission-critical tasks like network slicing in 6G. 📄 In our new paper, “LLM-Based Agentic Negotiation for 6G: Addressing Uncertainty Neglect and Tail-Event Risk,” we reveal how current AI agents (powered by Large Language Models) often ignore extreme events — the so-called “tail risks.” That’s a major problem in telecom where a single missed deadline can bring down a service. 🌪️ The problem: Many AI systems only focus on “average” outcomes. But in telecom, it's not the average that breaks your system, it’s the rare but catastrophic events. ✅ Our solution: A new risk-aware agentic AI framework that trains AI agents not just to optimize performance but to protect against worst-case scenarios. We combine: 📊 Digital Twins to simulate full latency distributions 🧠 Conditional Value at Risk (CVaR) to reason over worst-case outcomes 🔍 Epistemic Uncertainty Awareness so agents know how confident they are in their own decisions 💡 Why it matters: In tests involving eMBB and URLLC slices, our method eliminated SLA violations and reduced worst-case latencies by up to 11%, even when trading off some energy savings. In other words, more reliable networks, not just “smarter” ones. 🎯 This work is a step toward trustworthy autonomous 6G systems, where AI doesn’t just guess well, it reasons safely. 👏 Great work with all the team members: Hatim Chergui, Farhad Rezazadeh, and Mehdi Bennis. 💻 The code is available here: 👉 https://lnkd.in/d5eb2dya #6G #AI #AgenticAI #Telecom #LLM #NetworkSlicing #DigitalTwins #AIethics #OpenSource #PolynomeAI

Telecom Paper of the Day:  Why "Tiny" Logic Models Could Define the Future of Edge AI and 6G The prevailing dogma in #AI these days is "bigger is better", with #LLMs scaling into the trillions of parameters. However, a new pre-print published last month by Alexia Jolicoeur-Martineau challenges this assumption - and the implications for the telecommunications industry are potentially significant. A new "Tiny Recursive Model" (TRM) has outperformed massive LLMs on the ARC-AGI logic benchmark. The trick behind this feat lies in its architecture: Unlike LLMs that predict the next token based on massive memorization, the TRM utilizes a recursive, brain-inspired architecture. The model is capable of operating with just 7 million parameters - around 10,000 times smaller than frontier LLMs - and is achieving superior reasoning capabilities in specific logic tasks. For the telecom sector, the reliance on massive, power-hungry LLMs presents scalability and OpEx challenges. The TRM approach offers a potential roadmap for the next generation of Network Intelligence: ➡️ True Edge Intelligence: With a footprint of only 7M parameters, similar logic-based models could be deployed directly onto Customer Premises Equipment, RAN Intelligent Controllers, or even individual IoT sensors. This enables complex decision-making at the edge without the latency or bandwidth cost of a cloud round-trip. ➡️ Network Topology & Self-Optimization: The TRM excels at graph-based logic and puzzles. This mimics the mathematical complexity of network routing, fault isolation, and dynamic spectrum sharing - tasks that require strict logic rather than generative language capabilities. ➡️ Sustainability: As the world moves toward 6G, the energy efficiency of AI is critical. Moving away from brute-force compute toward recursive, specialized small models aligns with industry sustainability goals and reduces the hardware cost of AI integration. This research is a refreshing read for anyone interested: https://lnkd.in/efXmzyxZ Nature Magazine feature article: https://lnkd.in/eWm5D5n6 #Telecommunications #EdgeAI #ArtificialIntelligence #NetworkEngineering #6G #MachineLearning #SamsungAI #Innovation

🧠 Top AI Models Powering Next-Gen Telecom Networks Telecom networks generate more data every second than any human team could ever analyze. That’s exactly why AI models have become the intelligence layer of modern networks. From RAN KPIs and core logs to transport traces and real-time QoE indicators, AI is turning overwhelming telemetry into predictive signals, autonomous decisions, and self-optimizing operations. Here are the key AI models shaping the future of telecom 👇 🔹 Large Language Models (LLMs) Your new AI copilots, reading logs, summarizing alarm storms, validating configurations, generating scripts, and recommending likely fixes in natural language. 🔹 Anomaly Detection Models They learn the network’s “normal” and instantly flag micro-deviations like: fronthaul timing drifts, congestion fingerprints, packet-loss signatures, and early security threats. 🔹 Root Cause Analysis (RCA) Models These models correlate KPIs, topology dependencies, logs, and historical patterns to deliver instant root-cause hypotheses with recommended actions. 🔹 Reinforcement Learning (RL) The engine of true autonomy, optimizing spectrum allocation, beamforming, interference control, handovers, and UE mobility dynamically. 🔹 Forecasting & Predictive Analytics Models They predict demand surges, congestion zones, attenuation trends, power usage, and QoS degradation, enabling proactive network planning. 🔹 Graph Neural Networks (GNNs) They understand topology like a living map, modeling failure propagation, identifying bottlenecks, optimizing routing, and simulating RAN/CU/DU changes. 🔹 Computer Vision (CV) Models Powering autonomous field operations: drone inspections, fiber trench scoring, equipment labeling, improper installation detection, and RF heatmap analysis. 🌐 The Big Shift The future of telecom won’t be defined by bigger hardware, it will be defined by smarter networks. AI models are becoming the brain of telecom infrastructure, enabling networks that can: ✅ monitor themselves ✅ diagnose themselves ✅ optimize themselves ✅ and ultimately… heal themselves 🌍 Follow Abhishek Singh for visionary insights on how AI, automation, and next-gen networks are redefining global connectivity. #AI #Telecom #AIOps #NetworkAutomation #5G #6G #LLM #MachineLearning #DigitalTransformation #NetworkIntelligence #OpenRAN #EdgeComputing #SelfHealingNetworks #FutureOfNetworks #GenAI