Meta-Learning Biologically Plausible Plasticity Rules with Random Feedback Pathways
N. Shervani-Tabar, R. Rosenbaum
Journal: Nature Communications 14, no. 1 (2023): 1805
Backpropagation is widely used to train artificial neural networks, but its relationship to synaptic plasticity in the brain is unknown. Some biological models of backpropagation rely on feedback projections that are symmetric with feedforward connections, but experiments do not corroborate the existence of such symmetric backward connectivity. Random feedback alignment offers an alternative model in which errors are propagated backward through fixed, random backward connections. This approach successfully trains shallow models, but learns slowly and does not perform well with deeper models or online learning. In this study, we develop a novel meta-plasticity approach to discover interpretable, biologically plausible plasticity rules that improve online learning performance with fixed random feedback connections. The resulting plasticity rules show improved online training of deep models in the low data regime. Our results highlight the potential of meta-plasticity to discover effective, interpretable learning rules satisfying biological constraints.
[ Slides ] [ PDF ] [ Code ] |
Physics-Constrained Predictive Molecular Latent Space Discovery with Graph Scattering Variational Autoencoder
N. Shervani-Tabar, N. Zabaras
Preprint: arXiv preprint arXiv:2009.13878 (2020)
Conference: SIAM Conference on Computational Science and Engineering (CSE21), March 1–5, 2021, Fort Worth, Texas; Oral Presentation
Conference: SIAM Conference on Computational Science and Engineering (CSE21), March 1–5, 2021, Fort Worth, Texas; Oral Presentation
Recent advances in artificial intelligence have propelled the development of innovative computational materials modeling and design techniques. Generative deep learning models have been used for molecular representation, discovery, and design. In this work, we assess the predictive capabilities of a molecular generative model developed based on variational inference and graph theory in the small data regime. Physical constraints that encourage energetically stable molecules are proposed. The encoding network is based on the scattering transform with adaptive spectral filters to allow for better generalization of the model. The decoding network is a one-shot graph generative model that conditions atom types on molecular topology. A Bayesian formalism is considered to capture uncertainties in the predictive estimates of molecular properties. The model's performance is evaluated by generating molecules with desired target properties.
[ Slides ] [ PDF ] [ Code ]
[ Slides ] [ PDF ] [ Code ]
Multiresolution Approximation of a Bayesian Inverse Problem using Second-Generation Wavelets
N. Shervani-Tabar
Preprint: arXiv preprint arXiv:2304.00542 (2023)
Bayesian approaches are one of the primary methodologies to tackle an inverse problem in high dimensions. Such an inverse problem arises in hydrology to infer the permeability field given flow data in a porous media. It is common practice to decompose the unknown field into some basis and infer the decomposition parameters instead of directly inferring the unknown. Given the multiscale nature of permeability fields, wavelets are a natural choice for parameterizing them. This study uses a Bayesian approach to incorporate the statistical sparsity that characterizes discrete wavelet coefficients. First, we impose a prior distribution incorporating the hierarchical structure of the wavelet coefficient and smoothness of reconstruction via scale-dependent hyperparameters. Then, Sequential Monte Carlo (SMC) method adaptively explores the posterior density on different scales, followed by model selection based on Bayes Factors. Finally, the permeability field is reconstructed from the coefficients using a multiresolution approach based on second-generation wavelets. Here, observations from the pressure sensor grid network are computed via Multilevel Adaptive Wavelet Collocation Method (AWCM). Results highlight the importance of prior modeling on parameter estimation in the inverse problem.
[ Slides ] [ PDF ]
[ Slides ] [ PDF ]
Stabilized Conservative Level Set Method
N. Shervani-Tabar, O. V. Vasilyev
Journal: Journal of Computational Physics 375 (2018): 1033-1044
Conference: 69th Annual Meeting of the APS Division of Fluid Dynamics, November 20–22, 2016, Portland, Oregon; Oral Presentation
Conference: 69th Annual Meeting of the APS Division of Fluid Dynamics, November 20–22, 2016, Portland, Oregon; Oral Presentation
This research presents a Stabilized Conservative Level Set methodology. In order to conduct this research, a computational approach involving integration of conservative level set method with adaptive wavelet collocation method was developed to track the propagation of interfaces. A renormalization vector, m, has been implemented in the context of conservative level set method to substitute ill-defined normal vector, n. Also, an additional term has been added to stabilize the solution in regions far from the interface. Due to high gradient change on the front and low rate of change for variables in the regions far from the interface, Adaptive Mesh Refinement methods are a suitable choice for domain grid in the level set method. In study, Adaptive Wavelet Collocation Method has been used to demonstrate the methodology. This technique approaches to be a base to investigate the growth and collapse of a cavitation bubble and study its effects on cell membrane.
[ Slides ] [ PDF ] |
|
Experimental and Computational Investigation on the Cavitation Phenomenon in a Centrifugal Pump
N. Shervani-Tabar, R. Sedaaghi, M. T. Shervani-Tabar, R. Mohajerin, R. I. Bourisli
CAV 2012, August 13-16, 2012, Singapore, Singapore; Oral Presentation
First investigated experimentally, and then, simulated and computed by the CFD based software, FLUENT, this paper studies the effects of cavitation phenomenon within blades of a centrifugal pump. The casing of the centrifugal pump is partly made by Perspex, therefore the cavitation inception and development has been observed by a high-speed camera, assisted by a stroboscopic light. The centrifugal pump has been run in different suction pressures. In each case the flow rate and the pressure in the suction line of the pump have been slightly decreased by using a control valve. In every case, shots of cavitation phenomenon have been taken. Also pressure in the delivery side of the pump, pump speed, input power and output power of the pump have been measured and pump performance diagrams have been constructed. Therefore, performance of the pump can be analyzed in the cavitation-free working range of the pump as well as during inception and development of the cavitation phenomenon within blades of the pump. The experimental results have been compared with the computational ones. Results show that even a slight pressure drop in the suction line of the pump has a significant effect on the inception and development of the cavitation phenomenon within blades of the centrifugal pump.
[ Slides ] [ PDF ]
[ Slides ] [ PDF ]
Movement of Location of Tip Vortex Cavitation along Blade Edge due to Reduction of Flow Rate in an Axial Pump
M. T. Shervani-Tabar , N. Shervani-Tabar
Journal: International Journal of Mechanical and Aerospace Engineering, v6-33 (2012): 191-195
Conference: ICAMAME 2012, January 29-30, 2012, Dubai, UAE; Oral Presentation
Conference: ICAMAME 2012, January 29-30, 2012, Dubai, UAE; Oral Presentation
Tip vortex cavitation is one of well known patterns of cavitation phenomenon which occurs in axial pumps. This pattern of cavitation occurs due to pressure difference between the pressure and suction sides of blades of an axial pump. Since the pressure in the pressure side of the blade is higher than the pressure in its suction side, thus a very small portion of liquid flow flows back from pressure side to the suction side. This fact is cause of tip vortex cavitation and gap cavitation that may occur in axial pumps. In this paper the results of a series of experimental investigation on movement of tip vortex cavitation along blade edge due to reduction of pump flow rate in an axial pump is reported. Results show that reduction of pump flow rate in conjunction with increasing of outlet pressure causes movement of tip vortex cavitation along blade edge towards the blade tip. Results also show that by approaching tip vortex cavitation to the blade tip, vortex tip pattern of cavitation replaces with a cavitation phenomenon on the blade tip. Furthermore by further reduction of pump flow rate and increasing of outlet pressure, an unstable cavitation phenomenon occurs between each blade leading edge and the next blade trailing edge.
[ Slides ] [ PDF ]
[ Slides ] [ PDF ]