Design strategy transfer from humans to computers and across problems
Design problems involve planning and strategizing, where human designers identify meta-sub processes and routines of sequential actions relevant to the problem. This is an abstract skill that designers learn over time and then use across similar problems. This work presents an approach to represent design strategies using a probabilistic model. The model provides a mechanism to generate new designs based on certain design strategies. This work also demonstrates that this probabilistic representation can be used to transfer strategies from human designers to computational design agents in a way that is general and useful. This transfer-driven approach opens up the possibility of identifying high-performing behavior in human designers and using it to guide computational design agents. Also, a quintessential behavior of transfer learning is illustrated by agents, as transferring design strategies across different problems led to an improvement in agent performance. The agents in this work leverage the Cognitively Inspired Simulated Annealing Teams (CISAT) framework, an agent-based model that has been shown to mimic human problem-solving in configuration design problems. We apply this methodology on a cooling system design problem where the task is to determine the configuration of a cooling system in a given house layout. For the transfer learning experiments we use a medium size house to learn human strategies from and then use a larger and smaller version of this house layout as unseen problems. Following is a figure that shows the house layout of medium size (left) and large size (right).
The design agents use an online learning version of CISAT as the framework for making design decisions. CISAT incorporates Hidden Markov Models (HMMs) as a means of encoding probabilistic relationships between sequential actions. The online learning algorithm uses multi-arm bandit based formulation for increasing action probabilities that lead to positive rewards and reducing ones that lead to negative rewards. Previously studies have shown that these relationships can represent performance levels in human designers. Hence multiple experiments are executed with different initial values of these probabilistic relationships. The experiments were about finding trends when these strategies are extracted from human data but applied in a computational setting. The effect of these design strategies was observed on the performance of the agents. Different problems were used to evaluate the performance, one of the problems being the same as humans were asked to solve and two were new problems that were unseen in the human data, which was used to evaluate how effective the strategies were on an unseen but similar problem.
The work demonstrates the successful transfer of design strategies from human designers to computational agents. Following plot shows the comparative performance of the different strategies extracted from humans from the medium size house when applied on a large size house. The results show that design strategies from human designer data performed similar trends when applied by computational agents. Human design heuristics were successfully represented through probabilistic models and hence establish a common ground between human designers and computational agents for representing design strategies. These models leverage past human experiences using offline learning to augment the abilities of computational agents. Some experimental results also begin to explore this intersection of transfer learning and engineering design. The strategies are tested on three different problems, and certain human derived strategies learned from human data shows the best performance. An increase in agent performance is seen on different problems especially in the beginning of the design process, and this signifies that using previous experience in the form of design strategies helped agents perform better and faster than a randomly initialized agent. These inferences illustrate how previous design experiences could be used to develop generalizable strategies and improve performance across new design tasks.
