It’s been 20 years since the concept of cognitive radio was proposed—radio that could automatically switch from wireless signal to wireless signal, to minimize congestion and allow for more concurrent wireless connections. While this technology has been key in implementing intelligent communications, it’s even more interesting to look at the way cognition has developed into artificial intelligence. That’s exactly what the team working on the paper “20 Years of the Evolution from Cognitive to Intelligent Communications” wanted to see. The team, comprised of Zhijin Qin, Yue Gao (both from Queen Mary University of London), Xiangwei Zhou (of Louisiana State University), Lin Zhang, Ying-Chang Liang (both from the University of Electronic Science and Technology of China), and Geoffrey Ye Li (of the Georgia Institute of Technology), looked at the history, future, and challenges of making this development.
[Related Article: Best Releases and Papers from OpenAI in 2019 So Far]
Cognitive radio was proposed by J. Mitola and G.Q. Maguire in the paper “Cognitive radio: making software radios more personal” using a now-famous perception-action cycle: the process of intelligent decision-making. Using this, the cognitive devices are perceiving the parameters from an environment, applying them to the radio environment, making decisions based on that information, and then starting again with the new information from that decision. This all enhances spectrum utilization, since the devices are able to learn and adapt to what is needed.
The future of intelligent decision-making processes really relies on the change from using a cognitive agent to an AI agent—AI agents are more powerful and have higher capabilities. The paper discusses three reasons:
- “The AI agent has a better generalization functionality than the cognitive agent.” AI is able to make good and predictable decisions even if it has incomplete and inaccurate information, because it can learn more about the environment overall.
- The AI agent is able to predict better than the cognitive agent: “it can track the variation pattern of the radio environment and infer a proper action”
- The AI agent has better reasoning functionality, compared to the cognitive agent. It can learn the impact of an action on the environment quickly, rather than having to engage with complicated mathematical formulations.
The paper closes by addressing a few challenges that this technology may face, moving forward. These include:
- AI-based intelligent communications often use open data sets, which are difficult to control for quality or validity;
- They’ve been using different machine learning or deep learning frameworks, rather than tailor-made ones for this subject, meaning they aren’t as efficient as they could be and may not be interpretable;
- And they’re still learning to balance intelligence versus reliability: “some abnormal events might mislead the intelligent systems, which could further guide the whole system into a status that makes wrong or even unacceptable decisions”
[Related Article: The Best Machine Learning Research of Summer 2019]
Overall, there’s a lot of work which has been done with intelligent communication, but there’s still a lot to be done in the future.