• Corpus ID: 245769695

Pruning deep neural networks generates a sparse, bio-inspired nonlinear controller for insect flight

@inproceedings{Zahn2022PruningDN,
  title={Pruning deep neural networks generates a sparse, bio-inspired nonlinear controller for insect flight},
  author={Olivia Zahn and Jorge Luis Bustamante and Callin M. Switzer and Thomas Daniel and J. Nathan Kutz},
  year={2022}
}
Insect flight is a strongly nonlinear and actuated dynamical system. As such, strategies for understanding its control have typically relied on either model-based methods or linearizations thereof. Here we develop a framework that combines model predictive control on an established flight dynamics model and deep neural networks (DNN) to create an efficient method for solving the inverse problem of flight control. We turn to natural systems for inspiration since they inherently demonstrate… 

Figures and Tables from this paper

Data-driven sensor placement with shallow decoder networks
TLDR
Two algorithms for optimizing sensor locations for use with shallow decoder networks are developed: one which is a linear selection algorithm based upon QR (Q-SDN), and one which are a nonlinear selection algorithms based upon neural network pruning (P- SDN), which promise to enhance the already impressive reconstruction capabilities of SDNs.

References

SHOWING 1-10 OF 42 REFERENCES
Deep model predictive flow control with limited sensor data and online learning
TLDR
A novel deep learning model predictive control framework is presented that exploits low-rank features of the flow in order to achieve considerable improvements to control performance, using a recurrent neural network to accurately predict the control relevant quantities of the system.
Scalable training of artificial neural networks with adaptive sparse connectivity inspired by network science
TLDR
A method to design neural networks as sparse scale-free networks, which leads to a reduction in computational time required for training and inference, which has the potential to enable artificial neural networks to scale up beyond what is currently possible.
Abdominal movements in insect flight reshape the role of non-aerodynamic structures for flight maneuverability I: Model predictive control for flower tracking
Research on insect flight control has focused primarily on the role of wings. Yet abdominal deflections during flight can potentially influence the dynamics of flight. This paper assesses the role of
Deep learning for universal linear embeddings of nonlinear dynamics
TLDR
It is often advantageous to transform a strongly nonlinear system into a linear one in order to simplify its analysis for prediction and control, so the authors combine dynamical systems with deep learning to identify these hard-to-find transformations.
Flexible strategies for flight control: an active role for the abdomen
TLDR
A novel mechanism by which articulation of the body or ‘airframe’ of an animal can be used to redirect lift forces for effective flight control is suggested, and the small stability margin may increase flight agility by easing the transition from stable flight to a more maneuverable, unstable regime.
Learning Sparse Networks Using Targeted Dropout
TLDR
Target dropout is introduced, a method for training a neural network so that it is robust to subsequent pruning, and improves upon more complicated sparsifying regularisers while being simple to implement and easy to tune.
Learning Sparse Neural Networks through L0 Regularization
TLDR
A practical method for L_0 norm regularization for neural networks: pruning the network during training by encouraging weights to become exactly zero, which allows for straightforward and efficient learning of model structures with stochastic gradient descent and allows for conditional computation in a principled way.
The aerodynamics of insect flight
  • S. Sane
  • Biology
    Journal of Experimental Biology
  • 2003
TLDR
The basic physical principles underlying flapping flight in insects, results of recent experiments concerning the aerodynamics of insect flight, as well as the different approaches used to model these phenomena are reviewed.
Group sparse regularization for deep neural networks
Flight control in the hawkmoth Manduca sexta: the inverse problem of hovering
TLDR
The multiplicity of possible hovering kinematics shows that the means by which Manduca sexta actually maintains position and orientation may have considerable freedom and therefore may be influenced by many other factors beyond the physical and aerodynamic requirements of hovering flight.
...
...