This project aims at developing a novel computational framework that can adaptively re-use the knowledge across multiple machines potentially communicating via internet, to enhance the productivity in problem-solving. The developed framework comprises a class of optimization algorithms which are capable of automatically transferring knowledge across problem-solving tasks and by which the complicated real-world problems can be efficiently tackled. Different from the optimizers in the existing multitudinous Bayesian frameworks which are computationally demanding and prevailingly applicable in continuous search space, the optimization algorithms considered by the proposed framework are constraint-free in search space, and rely on nature-inspired, derivative-free, population-based search strategies which require little domain expertise on the user’s side. Besides, the proposed framework is significantly flexible in model selection and combination, and is highly parallelizable due to its emphasis on population-based search. The considered class of algorithms is regarded as a potential shortcut through which artificial general intelligence is achieved.
This project aims at developing a novel computational framework that can adaptively re-use the knowledge across multiple machines potentially communicating via internet, to enhance the productivity in problem-solving. The developed framework comprises a class of optimization algorithms which are capable of automatically transferring knowledge across problem-solving tasks and by which the complicated real-world problems can be efficiently tackled. Different from the optimizers in the existing multitudinous Bayesian frameworks which are computationally demanding and prevailingly applicable in continuous search space, the optimization algorithms considered by the proposed framework are constraint-free in search space, and rely on nature-inspired, derivative-free, population-based search strategies which require little domain expertise on the user’s side. Besides, the proposed framework is significantly flexible in model selection and combination, and is highly parallelizable due to its emphasis on population-based search. The considered class of algorithms is regarded as a potential shortcut through which artificial general intelligence is achieved.