Meta-learning algorithms leverage regularities that are present on a set of tasks to speed up and improve the performance of a subsidiary learning process. Recent work on deep neural networks has shown that prior gradient-based learning of meta-parameters can greatly improve the efficiency of subsequent learning. Here, we present a biologically plausible meta-learning algorithm based on equilibrium propagation. Instead of explicitly differentiating the learning process, our contrastive meta-learning rule estimates meta-parameter gradients by executing the subsidiary process more than once. This avoids reversing the learning dynamics in time and computing second-order derivatives. In spite of this, and unlike previous first-order methods, our rule recovers an arbitrarily accurate meta-parameter update given enough compute. We establish theoretical bounds on its performance and present experiments on a set of standard benchmarks and neural network architectures.
When an action potential arrives at a synapse there is a large probability that no neurotransmitter is released. Surprisingly, simple computational models suggest that these synaptic failures enable information processing at lower metabolic costs. However, these models only consider information transmission at single synapses ignoring the remainder of the neural network as well as its overall computational goal. Here, we investigate how synaptic failures affect the energy efficiency of models of entire neural networks that solve a goal-driven task. We find that presynaptic stochasticity and plasticity improve energy efficiency and show that the network allocates most energy to a sparse subset of important synapses. We demonstrate that stabilising these synapses helps to alleviate the stability-plasticity dilemma, thus connecting a presynaptic notion of importance to a computational role in lifelong learning. Overall, our findings present a set of hypotheses for how presynaptic plasticity and stochasticity contribute to sparsity, energy efficiency and improved trade-offs in the stability-plasticity dilemma.
Recent developments in the miniaturization of hardware have facilitated the use of robots or mobile sensory agents in many applications such as exploration of GPS-denied, hardly accessible unknown environments. This includes underground resource exploration and water pollution monitoring. One problem in scaling-down robots is that it puts significant emphasis on power consumption due to the limited energy available online. Furthermore, the design of adequate controllers for such agents is challenging as representing the system mathematically is difficult due to complexity. In that regard, Evolutionary Algorithms (EA) is a suitable choice for developing the controllers. However, the solution space for evolving those controllers is relatively large because of the wide range of the possible tunable parameters available on the hardware, in addition to the numerous number of objectives which appear on different design levels. A recently-proposed method, dubbed as Instinct Evolution Scheme (IES), offered a way to limit the solution space in these cases. This scheme uses Behavior Trees (BTs) to represent the robot behaviour in a modular, re-usable and intelligible fashion. In this paper, we improve upon the original IES by using Grammatical evolution (GE) to implement a full BT evolution model integratable with IES. A special emphasis is put on minimizing the complexity of the BT generated by GE. To test the scheme, we consider an environment exploration task on a virtual environment. Results show 85% correct reactions to environment stimuli and a decrease in relative complexity to 4.7%. Finally, the evolved BT is represented in an if-else on-chip compatible format.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.