2018/06/22 by Chrisantha Fernando, Chrisantha Thomas Fernando, Jakub Sygnowski +7 · 2 citations
Computer Science · Physics and Astronomy · #Adaptation (eye) #Artificial intelligence #Artificial neural network #Computer science #Hyperparameter #Machine learning #Meta learning (computer science) #Model Reduction and Neural Networks #Neural Networks and Applications #Neural Networks and Reservoir Computing #Reinforcement learning #Set (abstract data type) #Supervised learning #Task (project management) #cs.AI #cs.LG #cs.NE
paper · pdf · doi:10.1145/3205651.3205763
arxiv created 2018/06/22 · arxiv updated 2018/06/25 · openalex publication_date 2018/07/06 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
We show that the Baldwin effect is capable of evolving few-shot supervised and reinforcement learning mechanisms, by shaping the hyperparameters and the initial parameters of deep learning algorithms. This method rivals a recent meta-learning algorithm called MAML "Model Agnostic Meta-Learning," which uses second-order gradients instead of evolution to learn a set of reference parameters that can allow rapid adaptation to tasks sampled from a distribution. The Baldwin effect does not require gradients to be backpropagated to the reference parameters or hyperparameters, and permits effectively any number of gradient updates in the inner loop, learning strong learning dependent biases.