We propose a learning-optimisation paradigm to scientifically facilitate various design processes. The core idea is to propose a paradigm by making use of machine learning and optimization methods to encode, facilitate and automate the underlying design processes. The fundamental contribution will be its novel two-steps approach, which is adaptable, scalable and extensible over any ad-hoc rules-based or labeled interval calculus approach. The learning component is useful to capture and encode the abstract rules of the given design problem and the optimisation component allows efficient computation of multiple optimal solutions. It will be very much similar to the design principle of some actual design processes where the designers have their prior domain knowledge and sample between multiple design solutions in their brain based on their knowledge.

The power of the proposed paradigm lies in its generality to encode, facilitate and automate a wide spectrum of design problems such as interior design, fashion design, modular robot configuration, troop formation/planning or flocking and even sustainable city planning. Through applying the learning-optimisation approach on various problems, we will move ourselves closer to our ultimate goal, i.e., to develop a new design principle that could describe and answer the theoretical question: Why learning-optimisation approach is suitable or good for these kinds of problems. Though it is a difficult question to answer, it will be important to start by investigating the common features shared among the various problems. The proposed paradigm hence creates a computation platform for us to investigate new design principles.