...

• Intelligent networking is rapidly moving out of the lab and being deployed directly into production
• Operational maintenance, and service assurance are still a priority, but there is increasing interest in using AI/ML to drive network optimization and efficiency
• More research and development is needed to establish industry wide best practices and a shared understanding of intelligent networking to support interoperability.
• There has been some work on developing common or shared data sources and standards, but it remains a challenge
• LFN and the Open Source community are key contributors to furthering the development of intelligent networking now and in the future

Background

...

and History Beth Cohen 

0.5 page

At their hearts, telecoms are technology companies driven by the need to scale their networks to service millions of users, reliably, transparently and efficiently.   To achieve these ambitious goals, they need to optimize their networks by incorporating the latest technologies to feed the connected world's insatiable appetite for ever more bandwidth.  To do this efficiently, the networks themselves need to become more intelligent.  At the end of 2021, a bit over 2.5 years ago, LFN published its first white paper on the state of intelligent networking in the telecom industry.  Based on a survey of over 70 of its telecom community members, the findings pointed to a still nascent field made up of mostly research projects and lab experiments, with a few operational deployments related to automation and faster ticket resolutions.  The survey did highlight the keen interest of its respondents had in intelligent networking, machine learning and its promise for the future of the telecom industry in general. 

...

Data monetization must ensure data privacy, security and compliance with any existing regulations, and ensure that there is a trust that can be undeniable for those who provide their data under such regulations. This requires clear policies and security procedures to ensure safety at all times. This is a new strategic business opportunity for organizations looking to monetize anonymous data and leverage it for increased business efficiency and to determine product direction and determine new go to market strategies.

Intelligent Networking and AI Projects and Research

1.5 page

• l  Radio Access Network ChangJin Wang 
  • The intelligent evolution of wireless access networks is currently in a phase of rapid evolution and continuous innovation. In June 2022, 3GPP announced the freezing of R17 and described the process diagram of an intelligent RAN in TR37.817, including data collection, model training, model inference, and execution modules, which together form the infrastructure of an intelligent RAN. This promotes the rapid implementation and deployment of 5G RAN intelligence and provides support for intelligent scenarios such as energy saving, load balancing, and mobility optimization.
  • AI and Machine Learning Drive 5G RAN Intelligence

...

• l  Core Network Hui Deng ChangJin Wang 
  • Mobile core network is kind of central  or brain of mobile communication, it is the only and largest part which experienced the transformation from legacy proprietary hardware into telco cloud native, almost 100% of today mobile core network has been deployed based on telco cloud architecture by NFV technologies. It mostly consisted of packet core network such as 5GC and UPF which is responsible for packet forwarding, IMS which help operators multimedia communications such as voice, message and video communication of telecommunication operators, and the management of core network including telco cloud infrastructure and 5G network functionalities.  F.or those 3 parts of AI evolvement,

  • Network Intelligence Enables Experience Monetization and Differentiated Operations

    For a long period of time, operators have strived to realize traffic monetization on MBB(Mobile Broadband) networks. However, there are three technical gaps: not assessable user experience, no dynamic optimization, and no-closed-loop operations. To bridge these gaps,  there is strong requirement for Intelligent Personalized Experience solution, aiming to help operators add experience privileges to service packages and better monetize differentiated experiences. In the industry, the user plane on the mobile core network usually processes and forwards one service flow using one vCPU. As heavy-traffic services increase, such as 2K or 4K HD video and live streaming, microbursts and elephant flows frequently occur. It is, therefore, more likely that a vCPU will become overloaded, causing packet loss. To address this issue, Intelligent AI supported 5G core network can deliver ubiquitous 10 Gbps superior experiences.

  • Service Intelligence Expands the Profitability of Calling Services

    In 2023, New Calling was put into commercial use based on 3GPP specification, it could enhanced intelligence and data channel (3GPP specification)-based interaction capabilities; it is taking user to a multi-modal communication era and helping operators reconstruct their service layout. In addition, 3GPP architecture allow users to control digital avatars through voice during calls, delivering a more personalized calling experience. An enterprise can also customize their own avatar as an enterprise ambassador to promote their branding.

  • O&M Intelligence Achieves High Network Stability and Efficiency

    Empowered by the multi-modal large model, the Digital Assistant & Digital Expert (DAE) based AI technology could reduces O&M workload and improves O&M efficiency. It reshapes cloud-based O&M from "experts+tools" to intelligence-centric "DAE+manual assistance". With DAE, 80% of telecommunication operators trouble tickets can be automatically processed, which is much more efficient than 100% manual processing as it used to be. DAE also enables intent-driven O&M, avoiding manual decision-making. Before, it usually took over five years to cultivate experts in a single domain, however, the multi-modal large model is now able to be trained and updated within just weeks.

• l  Bearer & DC Network - Cisco?
• l  Cross domain network AI platform Lingli Deng //moved to 4.3.1
• l  Telecom Cloud Hui Deng 

issues solved/mitigated Lingli Deng Andrei Agapi 

• The advent of transformer models and attention mechanisms [][], and the sudden popularity of ChatGPT [], LLMs, transfer learning and foundation models in the NLP domain have all sparked vivid discussions and efforts to apply generative models in many other domains [].

  Interestingly, all of: word embeddings [], sequence models such as LSTMs [] and GRUs [], attention mechanisms [], transformer models [] and pretrained LLMs [][] have long been around before the launch of the ChatGPT tool in late 2022. Pretrained transformers like BERT[] in particular (especially transformer-encoder models) were very popular and widely used in NLP for tasks like sentiment analysis, text classification [], extractive question answering [] etc, long before ChatGPT made chatbots and decoder-based generative models go viral.

• That said, there has clearly been a spectacular explosion of academic research, commercial activity and ecosystems that have emerged since ChatGPT came out, in the area of both open [][][] and closed source [][][] LLM foundation models, related software, services and training datasets.

  Beyond typical chatbot-style applications, LLMs have been extended to generate code [][], solve Math problems (stated either formally or informally) [], pass science exams [][], or act as incipient "AGI"-style agents for different tasks, including advising on investment strategies, or setting up a small business [][]. Recent advancements to the basic LLM text generation model include instruction finetuning [], retrieval augmented generation using external vector stores [][], using external tools such as web search [], external knowledge databases or other APIs for grounding models [], code interpreters, calculators and formal reasoning tools [][]. Beyond LLMs and NLP, transformers have also been used to handle non-textual data, such as images [], sound [] and arbitrary sequence data [].

• A natural question arises on how the the power of LLMs can be harnessed for problems and applications related to Intelligent Networking, network automation and for operating and optimizing telecommunication networks in general, at any level of the network stack.

  Datasets encountered in telco-related applications have a few particularities. For one, data one might encounter ranges from fully structured (e.g. code, scripts, configuration, or time series KPIs), to semi-structured (syslogs, design templates etc), to unstructured data (design documents and specifications, Wikis, Github issues, emails, chatbot conversations).

• Another issue is domain adaptation. Language encountered in telco datasets can be very domain specific (including CLI commands and CLI output, formatted text, network slang and abbreviations, syslogs, RFC language, network device specifications etc). Off-the-shelf performance of LLM models strongly depends on whether those LLMs have actually seen that particular type of data during training (this is true for both generative LLMs and embedding models). There exist several approaches to achieve domain adaptation and downstream task adaptation of LLM models. In general these either rely on 1) In-context-learning, prompting and retrieval augmentation techniques; 2) Finetuning the models; or 3) Hybrid approaches. For finetuning LLMs, unlike for regular neural network models, several specialized techniques exist in the general area of PEFT (Parameter Efficient Fine Tuning), allowing one to only finetune a very small percentage of the many billions of parameters of a typical LLM. In general, the best techniques to achive domain adaptation for an LLM will heavily depend on: 1) the kind of data we have and how much domain data we have available, 2) the downstream task, and 3) the foundation model we start from. In addition to general domain adaptation, many telcos will have the issue of multilingual datasets, where a mix of languages (typically English + something else) will exist in the data (syslogs, wikis, tickets, chat conversations etc). While many options exist for both generative LLMs [] and text embedding models [], not many foundation models have seen enough non-English data in training, thus options in foundation model choice are definitely restricted for operators working on non-English data.

• In conclusion, while foundation models and transfer learning have been shown to work very well on general human language when pretraining is done on large corpuses of human text (such as Wikipedia, or the Pile[]), it remains an open question to be answered whether domain adaptation and downstream task adaptation work equally well on the kinds of domain-specific, semi-structured, mixed modality datasets we can find in telco networks. To enable this, telcos should very likely focus on standardization and data governance efforts, such as standardized and unified data collection policies and high quality structured data, as discussed earlier in this whitepaper.

• Deploying large models such as LLMs in production, especially at scale, also raises several other issues in terms of: 1) Performance, scalability and cost of inference, especially when using large context windows (most transformers scale poorly with context size); 2) Deployment of models in the cloud, on premise, multi-cloud, or hybrid; 3) Issues pertaining to privacy and security of the data for each particular application; 4) Issues common to many other ML/AI applications, such as ML-Ops, continuous model validation and continuous re-training.

• high quality structured data Large language models can be used to understand large amounts of unstructured operation and maintenance data (for example, system logs, operation and maintenance work orders, operation guides, company documents, etc., which are traditionally used in human-computer interaction or human-to-human collaboration scenarios), from which effective knowledge is extracted to provide guidance for further automatic/intelligent operation and maintenance, thereby effectively expanding the scope of the application of autonomous mechanism.
• AI trustworthiness moved to 4.3.2
• non-economic margin cost Although equipment manufacturers can provide many domain AI solutions for professional networks/single-point equipment, these solutions are limited in "field of view" and cannot solve problems that require a "global view" such as end-to-end service quality assurance and rapid response to faults. . Operators can aggregate management and maintenance data in various network domains by building a unified data sharing platform, and based on this, further provide a unified computing resource pool, basic AI algorithms and inference platform (i.e. cross-domain AI platform) for various scenario-specific AI for end-to-end scenarios and intra-domain scenarios. Applied reasoning platform.
• TBA

issues remained/added/aggravated Lingli Deng 

• AI trustworthiness
• non-economic margin cost Compared with traditional data-driven dedicated AI/ML models in specific scenarios, the R&D and operation of large language models has higher requirements for pre-training data scale, training computing power cluster scale, fine-tuning engineering and other resource requirements, energy consumption, management and maintenance, etc. It is bound to become a playground for the few. How to open up the R&D, application, operation and maintenance chain of large models for the communications network industry so that it can serve the general public may become a stepping stone for operators to realize high-end autonomous networks.
• TBA

Network LLM:

...

The game changer?

Jason Hunt (at least on how foundation models can be applied to network data) Andrei Agapi 

...