Meta-learning by the baldwin effect

@article{Fernando2018MetalearningBT,
  title={Meta-learning by the baldwin effect},
  author={Chrisantha Fernando and Jakub Sygnowski and Simon Osindero and Jane X. Wang and T. Schaul and Denis Teplyashin and P. Sprechmann and A. Pritzel and Andrei A. Rusu},
  journal={Proceedings of the Genetic and Evolutionary Computation Conference Companion},
  year={2018}
}
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… Expand
Meta-learnt priors slow down catastrophic forgetting in neural networks
TLDR
It is shown that catastrophic forgetting can be mitigated in a meta-learning context, by exposing a neural network to multiple tasks in a sequential manner during training. Expand
Why to "grow" and "harvest" deep learning models?
TLDR
It is shown that the stochastic gradient descent (SGD) with two balanced pre-defined values of per capita growth and harvesting rates outperform the most common adaptive gradient methods in all of the three requirements. Expand
Towards continual reinforcement learning through evolutionary meta-learning
TLDR
The introduced approach combines neuroevolution, to determine the starting weights of a deep neural network, and a version of deep reinforcement learning that is continually running during execution time, showing that the combined approach gives a better generalisation performance when compared to evolution or reinforcement learning alone. Expand
Meta-learners' learning dynamics are unlike learners'
TLDR
It is shown that, once meta-trained, LSTM Meta-Learners aren't just faster learners than their sample-inefficient deep learning and reinforcement learning brethren, but that they actually pursue fundamentally different learning trajectories. Expand
Evolving Self-supervised Neural Networks: Autonomous Intelligence from Evolved Self-teaching
  • Nam Le
  • Computer Science
  • ArXiv
  • 2019
TLDR
Experimental results show that only evolved self-supervised agents can demonstrate some sort of intelligent behaviour, but not evolution or self-learning alone. Expand
Meta-Teaching : Curriculum Generation for Lifelong Learning
  • 2019
Meta-learning will be crucial to creating lifelong, generalizable AI. In practice, however, it is hard to define the meta-training task distribution that is used to train meta-learners. If made tooExpand
Evolving intrinsic motivations for altruistic behavior
TLDR
It is demonstrated that individual inductive biases for cooperation can be learned in a model-free way by combining MARL with appropriately structured natural selection and an innovative modular architecture for deep reinforcement learning agents which supports multi-level selection. Expand
Social Learning vs Self-teaching in a Multi-agent Neural Network System
TLDR
This paper compares the effect on evolution of social learning in comparison with that of individual learning, and proposes a neural architecture called a “self-taught neural network” in order to allow an agent to learn on its own, without any supervision. Expand
On the Potential for Open-Endedness in Neural Networks
TLDR
This article reviews common barriers to open-endedness in the evolution-inspired approach and how they are dealt with in the evolutionary case and shows how these problems map onto similar ones in the machine learning approach, and discusses how the same insights and solutions that alleviated those barriers in evolutionary approaches can be ported over. Expand
ES-MAML: Simple Hessian-Free Meta Learning
TLDR
It is shown how ES can be applied to MAML to obtain an algorithm which avoids the problem of estimating second derivatives, and is also conceptually simple and easy to implement. Expand
...
1
2
...

References

SHOWING 1-10 OF 30 REFERENCES
Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
We propose an algorithm for meta-learning that is model-agnostic, in the sense that it is compatible with any model trained with gradient descent and applicable to a variety of different learningExpand
Optimization as a Model for Few-Shot Learning
Asynchronous Methods for Deep Reinforcement Learning
TLDR
A conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers and shows that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input. Expand
Learning to reinforcement learn
TLDR
This work introduces a novel approach to deep meta-reinforcement learning, which is a system that is trained using one RL algorithm, but whose recurrent dynamics implement a second, quite separate RL procedure. Expand
The baldwin effect in developing neural networks
TLDR
This research uses new evidence of the blurred distinction between development and learning in natural neural systems as the basis for an abstract model displaying the Baldwin Effect in artificial neural networks that evolve, develop and learn. Expand
One-shot Learning with Memory-Augmented Neural Networks
TLDR
The ability of a memory-augmented neural network to rapidly assimilate new data, and leverage this data to make accurate predictions after only a few samples is demonstrated. Expand
RL$^2$: Fast Reinforcement Learning via Slow Reinforcement Learning
TLDR
This paper proposes to represent a "fast" reinforcement learning algorithm as a recurrent neural network (RNN) and learn it from data, encoded in the weights of the RNN, which are learned slowly through a general-purpose ("slow") RL algorithm. Expand
The revival of the Baldwin effect
TLDR
It is argued that what was actually shown is that the learning ability is easily selected for, and it is found that learning can increase the chance of fixation of the learned behavior by several orders of magnitude compared with the non-learning situation. Expand
Siamese Neural Networks for One-Shot Image Recognition
TLDR
A method for learning siamese neural networks which employ a unique structure to naturally rank similarity between inputs and is able to achieve strong results which exceed those of other deep learning models with near state-of-the-art performance on one-shot classification tasks. Expand
Lamarckian Evolution and the Baldwin Effect in Evolutionary Neural Networks
TLDR
Two ways of combining learning and genetic search are explored: one exploits the Baldwin effect, while the other uses a Lamarckian strategy that gets a lower average error than other BP-based algorithms like RPROP. Expand
...
1
2
3
...