ML models and intelligent automation: the key to decentralized learning

Collaborative intelligence has been essential to the success of our learning and is now proving essential to ML models.

Human technological growth has always depended on collaborative intelligence. We haven’t just taken advantage of the data around us; we shared our discoveries with others, worked together to solve problems, and even became picky about who we learned from or exchanged our knowledge and expertise with. These were critical to the success of our learning and are now proving equally important in ML models for mobile networks. Next-generation autonomous cellular operators will be complex ecosystems composed of a large number of decentralized, intelligent network devices and network elements of nodes enabling decentralized learning, capable of simultaneously producing and distributing data using ML models and intelligent automation.

Distributed and decentralized learning techniques

Distributed ML approaches are considered most appropriate in a complex ecosystem of network components and devices, where data is inherently distributed and can be confidential and large. These strategies enable collaborative learning algorithms without requiring raw data exchange and can be used to integrate all local learning from inherently decentralized local datasets into a single unified ML model. This co-trained machine learning model can in turn help staff operate more efficiently through proactive outage management methods, improving both the quality of experience and operator revenue.

Decentralized learning and collaborative artificial intelligence (AI) will enable rapid training with less IT and network resource allocation and also improved efficiency – reducing network footprint, communication overhead, exchange of knowledge and energy consumption.

Address heterogeneity

Decentralized datasets in distributed learning contexts are diverse because they are acquired from multiple nodes and devices, which themselves are often heterogeneous. They can have various features and content, and they can be sampled from various distros. A base station, for example, can measure transmit power and network throughput at the cellular level, while a device can measure its own position, timing and background, as well as the presentation quality of the implementation it runs. Since all of this information is useful and necessary for accurate early predictions, optimizations, and proactive failure prevention, it should be integrated into the jointly trained global ML model.

Some networks or devices may also report negative inputs that reduce model performance, either intentionally, as in the case of attacks, or unintentionally, as in the case of information distribution changes or sensor errors. Such scenarios can affect overall model accuracy while increasing training time, power consumption, and network link utilization. The problem of data heterogeneity, on the other hand, can be addressed by an autonomous and adaptable orchestration of the learning experience.

Horizontal federated learning (HFL)

HFL allows the formation of a combined global model composed of various data samples on the same observation variables – in this case, ML features. We can perform local simulation of workers based on their own local datasets as they have both the input features (x) and the output labels (y). Since all compute nodes and the master share the same system model, the entire model is sharable and aggregable.

Split learning (SL)

When the decentralized network nodes have distinct ML features on the same instance of temporal data, SL allows the construction of an overall model. In this case worker nodes, for example, can only contain the input characteristics (X). Only the master server has access to the appropriate label (y). Therefore, working networks may have only part of the neural network model. Also, worker models do not need to have the same neural layers.

MAB agents in distributed machine learning

It can be difficult to predict which workers in a federation will benefit from the global model and which will jeopardize it. In the use case we studied, one of the worker nodes reported erroneous data to the master server – very different values ​​than the majority of worker nodes in the union. This scenario can have a severe influence on the jointly trained modeling framework, hence the need for a detection mechanism at the master server level to identify the malicious worker and prevent him from joining the federation. In this way, we hope to maintain the performance of the global model, in which the majority of workers continue to benefit from the federation despite the malicious contribution of some workers.

Therefore, when there is at least one employee in the federation who has a negative impact on the federation, it is essential to exclude this employee from global model changes. The previous methods depend on pre-hoc clustering, which does not allow near real-time modification. In this situation, we use MAB-based support to help the master server remove all unauthorized worker nodes.

More trending stories:

Share this article

Do the sharing