Skip to main content

Leading UK photonics company licenses PhD student’s deep learning technology

Research by TRANSNET PhD student with world-renowned industry partner finds real-world application.


EEE’s Optical Networks Group (ONG) student, Zacharaya Shabka, has developed a deep reinforcement learning (DRL) for fast and effective control parameter optimization under the supervision of Professor Georgios Zervas.

UCL Business (UCLB), the commercialization company for University College London (UCL), has entered into a commercial licensing agreement with Polatis for the use of the DRL technology, developed directly as a result of Zacharaya’s PhD project. 

UCLB has also filed a patent application for the method developed by Zacharaya to generate optimized control parameters in a production scalable manner.

Apoorva Srikkanth, Business Manager at UCLB states: 

"Zacharaya’s work with Polatis and the subsequent licensing outcome is an excellent example of how early engagement with industry partners allows for real-world impact of academic research. The academic team involved UCLB as soon as they saw a commercial potential in the project which helped us have early discussions with Polatis to develop a feasible technology transfer plan."

Working with UCLB, the academic team of Zacharaya and Prof. Zervas have successfully filed to patent the technology from the project and have licensed it alongside code written by Zacharaya. 

Prof Zervas - supervisor to Zacharaya and leading academic in the field of optical communications at UCL - states:

"I have closely collaborated with Polatis for over 15 years. This is one of the most impactful joint projects that had an immediate commercial value and adoption. It reflects the vision and ambition of my research team that works on systematic and disruptive research solutions across optical networked systems."

 


From research to license 

During the collaborative research project with Polatis, Zacharaya developed a method for optimising control parameters based on deep reinforcement learning (DRL). The method allows for lengthy tuning processes to be avoided entirely, instead generating parameters directly from some simple measurements of the underlying control system.

Control parameters can then be generated in milliseconds without requiring any further operation of the device. This feature makes the method highly efficient in mass-manufacturing scenarios, since optimisation time is minimal and can be implemented in parallel with other production processes.

Polatis’ software-defined optical circuit switches create highly transparent connections between hundreds of fibres and are used across numerous domains worldwide from secure communications to data centre networks.

The commercialisation of this DRL technology will allow Polatis to eliminate control optimisation time in production, as well as to realise improved optical switching performance and increased manufacturing efficiency.

Nick Parsons, Vice President Technology, HUBER+SUHNER, states:

"By engaging with UCL, we are able to multiply the effectiveness of our relatively small engineering team by understanding the impact of new technologies and exploiting them earlier than we would otherwise have done. Zacharaya’s novel deep learning method has shown promising results in increasing efficiency and yield in our manufacturing process and in improving product performance."

Zacharaya is currently consulting through UCL Consultants (UCLC) with Polatis to support the implementation of this method. The coordination between UCL Business and UCL Consultants has ensured that the IP will be put to good use and find tangible real-world benefit in a global manufacturing system.

The commercial outputs of Zacharaya’s work also highlight the real-world impact of the research occurring in EEE at UCL and the importance of partnership and co-creation between academia and industry.

The collaborative project was funded by an EPSRC CASE award. The award promotes the collaboration of industry and academia by providing a challenging research training experience, within the context of a mutually beneficial research collaboration with industry. UCL EEE prides itself on its strong portfolio of collaborations with industry.

The DRL method to optimise control parameters is applicable to other mass manufacturing industries and is available to license from UCLB.


Article Credit

The original article was published by UCL Electronic and Electrical Engineering and can be accessed here