arXiv subjects

Jinn-Liang Liu

Publications and source records attributed to Jinn-Liang Liu.

14 recordsLinked to original sources

arXiv2023

Multi-task UNet architecture for end-to-end autonomous driving

We propose an end-to-end driving model that integrates a multi-task UNet (MTUNet) architecture and control algorithms in a pipeline of data flow from a front camera through this model to driving decisions. It provides quantitative measures to evaluate the holistic, dynamic, and real-time performance of end-to-end driving systems and thus the safety and interpretability of MTUNet. The architecture consists of one segmentation, one regression, and two classification tasks for lane segmentation, path prediction, and vehicle controls. We present three variants of the architecture having different complexities, compare them on different tasks in four static measures for both single and multiple tasks, and then identify the best one by two additional dynamic measures in real-time simulation. Our results show that the performance of the proposed supervised learning model is comparable to that of a reinforcement learning model on curvy roads for the same task, which is not end-to-end but multi-module.

cs.LG↗

arXiv2022

Prediction of activity coefficients in water-methanol mixtures using a generalized Debye-Huckel model

We propose a generalized Debye-Huckel model from Poisson-Fermi theory to predict the mean activity coefficient of electrolytes in water-methanol mixtures with arbitrary percentage of methanol from 0 to 100%. The model applies to any number of ionic species and accounts for both short and long ion-ion, ion-water, ion-methanol, and water-methanol interactions, the size effect of all particles, and the dielectric effect of mixed-solvent solutions. We also present a numerical algorithm with mathematical and physical details for using the model to fit or predict experimental data. The model has only 3 empirical parameters to fit the experimental data of NaF, NaCl, and NaBr, for example, in pure-water solutions. It then uses another 3 parameters to predict the activities of these salts in mixed-solvent solutions for any percentage of methanol. Values of these parameters show mathematical or physical meaning of ionic activities under variable mixing condition and salt concentration. The algorithm can automatically determine optimal values for the 3 fitting parameters without any manual adjustments.

cond-mat.stat-mech↗

arXiv2021

End-to-End Deep Learning of Lane Detection and Path Prediction for Real-Time Autonomous Driving

Inspired by the UNet architecture of semantic image segmentation, we propose a lightweight UNet using depthwise separable convolutions (DSUNet) for end-to-end learning of lane detection and path prediction (PP) in autonomous driving. We also design and integrate a PP algorithm with convolutional neural network (CNN) to form a simulation model (CNN-PP) that can be used to assess CNN's performance qualitatively, quantitatively, and dynamically in a host agent car driving along with other agents all in a real-time autonomous manner. DSUNet is 5.16x lighter in model size and 1.61x faster in inference than UNet. DSUNet-PP outperforms UNet-PP in mean average errors of predicted curvature and lateral offset for path planning in dynamic simulation. DSUNet-PP outperforms a modified UNet in lateral error, which is tested in a real car on real road. These results show that DSUNet is efficient and effective for lane detection and path prediction in autonomous driving.

cs.CV↗

arXiv2021

Differential Capacitance of Electric Double Layers: A Poisson-Bikerman Formula

We propose a Poisson-Bikerman (PBik) formula for calculating the differential capacitance (DC) of electrical double layers (EDLs) in aqueous electrolytes or ionic liquids. The PBik theory is a generalization of the classical Poisson-Boltzmann theory to include different steric energies of different-sized ions and water similar to different electrical energies for different-charged ions. Water and ions with interstitial voids in this molecular mean field theory have their physical volumes as they do in molecular dynamics simulations. The PBik formula derived from Fermi distributions of ions and water in arbitrary shape and volume reduces to the Bikerman-Freise formula derived from the lattice model of equal-sized ions. The DC curves predicted by the Gouy-Chapman formula are U-shaped (for point-like ions with zero volume and very dilute solutions). The curves change from U shape to camel shape (Bactrian) and then to bell shape (for finite size ions) as the volume fraction of ions and water changes from zero to medium value then to large value. The transition is characterized by critical and inflection voltages in terms of the particle volume fraction. These voltages determine steric and electrical energies that describe the space/charge competition and saturation properties of ions and water packed in the condensed layer of EDLs under high field conditions. Steric energy is as important as electrical energy in these conditions. PBik computes symmetric DC curves from delicately balanced steric interactions of asymmetric-size ions and water like the experimental data of KPF6 in aqueous solution. It computes asymmetric curves and captures delicately balanced steric or electrical interactions of ions having different volumes or charges in ionic liquids.

cond-mat.soft↗

arXiv2020

Molecular Mean-Field Theory of Ionic Solutions: a Poisson-Nernst-Planck-Bikerman Model

We have developed a molecular mean-field theory -- fourth-order Poisson-Nernst-Planck-Bikerman theory -- for modeling ionic and water flows in biological ion channels by treating ions and water molecules of any volume and shape with interstitial voids, polarization of water, and ion-ion and ion-water correlations. The theory can also be used to study thermodynamic and electrokinetic properties of electrolyte solutions in batteries, fuel cells, nanopores, porous media including cement, geothermal brines, the oceanic system, etc. The theory can compute electric and steric energies from all atoms in a protein and all ions and water molecules in a channel pore while keeping electrolyte solutions in the extra- and intracellular baths as a continuum dielectric medium with complex properties that mimic experimental data. The theory has been verified with experiments and molecular dynamics data from the gramicidin A channel, L-type calcium channel, potassium channel, and sodium/calcium exchanger with real structures from the Protein Data Bank. It was also verified with the experimental or Monte Carlo data of electric double-layer differential capacitance and ion activities in aqueous electrolyte solutions. We give an in-depth review of the literature about the most novel properties of the theory, namely, Fermi distributions of water and ions as classical particles with excluded volumes and dynamic correlations that depend on salt concentration, composition, temperature, pressure, far-field boundary conditions etc. in a complex and complicated way as reported in a wide range of experiments. The dynamic correlations are self-consistent output functions from a fourth-order differential operator that describes ion-ion and ion-water correlations, the dielectric response (permittivity) of ionic solutions, and the polarization of water molecules with a single correlation length parameter.

cond-mat.soft↗

arXiv2019

Deep Learning and Control Algorithms of Direct Perception for Autonomous Driving

Based on the direct perception paradigm of autonomous driving, we investigate and modify the CNNs (convolutional neural networks) AlexNet and GoogLeNet that map an input image to few perception indicators (heading angle, distances to preceding cars, and distance to road centerline) for estimating driving affordances in highway traffic. We also design a controller with these indicators and the short-range sensor information of TORCS (the open racing car simulator) for driving simulated cars to avoid collisions. We collect a set of images from a TORCS camera in various driving scenarios, train these CNNs using the dataset, test them in unseen traffics, and find that they perform better than earlier algorithms and controllers in terms of training efficiency and driving stability. Source code and data are available on our website.

cs.LG↗

arXiv2018

Analysis of Generalized Debye-Hückel Equation from Poisson-Fermi Theory

The Debye-Hückel equation is a fundamental physical model in chemical thermodynamics that describes the free energy (chemical potential, activity) of an ion in electrolyte solutions at variable salt concentration, temperature, and pressure. It is based on the linear Poisson-Boltzmann equation that ignores the steric (finite size), correlation, and polarization effects of ions and water (or solvent molecules). The Poisson-Fermi theory developed in recent years takes these effects into account. A generalized Debye-Hückel equation is derived from the Poisson-Fermi theory and is shown to consistently reduce to the classical equation when these effects vanish in limiting cases. As a result, a linear fourth-order Poisson-Fermi equation is presented for which unique solutions are shown to exist for spherically symmetric systems. Moreover, a generalized Debye length is proposed to include the size effects of ions and water.

physics.chem-ph↗

arXiv2017

Poisson-Fermi Modeling of Ion Activities in Aqueous Single and Mixed Electrolyte Solutions at Variable Temperature

The combinatorial explosion of empirical parameters in tens of thousands presents a tremendous challenge for extended Debye-Hückel models to calculate activity coefficients of aqueous mixtures of most important salts in chemistry. The explosion of parameters originates from the phenomenological extension of the Debye-Hückel theory that does not take steric and correlation effects of ions and water into account. In contrast, the Poisson-Fermi theory developed in recent years treats ions and water molecules as nonuniform hard spheres of any size with interstitial voids and includes ion-water and ion-ion correlations. We present a Poisson-Fermi model and numerical methods for calculating the individual or mean activity coefficient of electrolyte solutions with any arbitrary number of ionic species in a large range of salt concentrations and temperatures. For each activity-concentration curve, we show that the Poisson-Fermi model requires only three unchanging parameters at most to well fit the corresponding experimental data. The three parameters are associated with the Born radius of the solvation energy of an ion in electrolyte solution that changes with salt concentrations in a highly nonlinear manner.

cond-mat.soft↗

arXiv2017

A GPU Poisson-Fermi Solver for Ion Channel Simulations

The Poisson-Fermi model is an extension of the classical Poisson-Boltzmann model to include the steric and correlation effects of ions and water treated as nonuniform spheres in aqueous solutions. Poisson-Boltzmann electrostatic calculations are essential but computationally very demanding for molecular dynamics or continuum simulations of complex systems in molecular biophysics and electrochemistry. The graphic processing unit (GPU) with enormous arithmetic capability and streaming memory bandwidth is now a powerful engine for scientific as well as industrial computing. We propose two parallel GPU algorithms, one for linear solver and the other for nonlinear solver, for solving the Poisson-Fermi equation approximated by the standard finite difference method in 3D to study biological ion channels with crystallized structures from the Protein Data Bank, for example. Numerical methods for both linear and nonlinear solvers in the parallel algorithms are given in detail to illustrate the salient features of the CUDA (compute unified device architecture) software platform of GPU in implementation. It is shown that the parallel algorithms on GPU over the sequential algorithms on CPU (central processing unit) can achieve 22.8x and 16.9x speedups for the linear solver time and total runtime, respectively.

physics.comp-ph↗

arXiv2017

Poisson-Fermi Formulation of Nonlocal Electrostatics in Electrolyte Solutions

We present a nonlocal electrostatic formulation of nonuniform ions and water molecules with interstitial voids that uses a Fermi-like distribution to account for steric and correlation effects in electrolyte solutions. The formulation is based on the volume exclusion of hard spheres leading to a steric potential and Maxwell's displacement field with Yukawa-type interactions resulting in a nonlocal electric potential. The classical Poisson-Boltzmann model fails to describe steric and correlation effects important in a variety of chemical and biological systems, especially in high field or large concentration conditions found in and near binding sites, ion channels, and electrodes. Steric effects and correlations are apparent when we compare nonlocal Poisson-Fermi results to Poisson-Boltzmann calculations in electric double layer and to experimental measurements on the selectivity of potassium channels for K+ over Na+. The present theory links atomic scale descriptions of the crystallized KcsA channel with macroscopic bulk conditions. Atomic structures and macroscopic conditions determine complex functions of great importance in biology, nanotechnology, and electrochemistry.

cond-mat.soft↗

arXiv2016

A Nonlocal Poisson-Fermi Model for Ionic Solvent

We propose a nonlocal Poisson-Fermi model for ionic solvent that includes ion size effects and polarization correlations among water molecules in the calculation of electrostatic potential. It includes the previous Poisson-Fermi models as special cases, and its solution is the convolution of a solution of the corresponding nonlocal Poisson dielectric model with a Yukawa-type kernel function. Moreover, the Fermi distribution is shown to be a set of optimal ionic concentration functions in the sense of minimizing an electrostatic potential free energy. Finally, numerical results are reported to show the difference between a Poisson-Fermi solution and a corresponding Poisson solution.

physics.chem-ph↗

arXiv2015

Poisson-Fermi Model of Single Ion Activities

A Poisson-Fermi model is proposed for calculating activity coefficients of single ions in strong electrolyte solutions based on the experimental Born radii and hydration shells of ions in aqueous solutions. The steric effect of water molecules and interstitial voids in the first and second hydration shells play an important role in our model. The screening and polarization effects of water are also included in the model that can thus describe spatial variations of dielectric permittivity, water density, void volume, and ionic concentration. The activity coefficients obtained by the Poisson-Fermi model with only one adjustable parameter are shown to agree with experimental data, which vary nonmonotonically with salt concentrations.

physics.chem-ph↗

arXiv2015

Poisson-Nernst-Planck-Fermi Theory for Ion Channels

A Poisson-Nernst-Planck-Fermi (PNPF) theory is developed for studying ionic transport through biological ion channels. Our goal is to deal with the finite size of particle using a Fermi like distribution without calculating the forces between the particles, because they are both expensive and tricky to compute. We include the steric effect of ions and water molecules with nonuniform sizes and interstitial voids, the correlation effect of crowded ions with different valences, and the screening effect of water molecules in an inhomogeneous aqueous electrolyte. Including the finite volume of water and the voids between particles is an important new part of the theory presented here. Fermi like distributions of all particle species are derived from the volume exclusion of classical particles. The classical Gibbs entropy is extended to a new entropy form --- called Gibbs-Fermi entropy --- that describes mixing configurations of all finite size particles and voids in a thermodynamic system where microstates do not have equal probabilities. The PNPF model describes the dynamic flow of ions, water molecules, as well as voids with electric fields and protein charges. The PNPF results are in good accord with experimental currents recorded in a 10^8-fold range of Ca++ concentrations. The results illustrate the anomalous mole fraction effect, a signature of L-type calcium channels. Moreover, numerical results concerning water density, dielectric permittivity, void volume, and steric energy provide useful details to study a variety of physical mechanisms ranging from binding, to permeation, blocking, flexibility, and charge/space competition of the channel.

physics.bio-ph↗

arXiv2015

Numerical Methods for a Poisson-Nernst-Planck-Fermi Model

Numerical methods are proposed for an advanced Poisson-Nernst-Planck-Fermi (PNPF) model for studying ion transport through biological ion channels. PNPF contains many more correlations than most models and simulations of channels, because it includes water and calculates dielectric properties consistently as outputs. This model accounts for the steric effect of ions and water molecules with different sizes and interstitial voids, the correlation effect of crowded ions with different valences, and the screening effect of polarized water molecules in an inhomogeneous aqueous electrolyte. The steric energy is shown to be comparable to the electrical energy under physiological conditions, demonstrating the crucial role of the excluded volume of particles and the voids in the natural function of channel proteins. Water is shown to play a critical role in both correlation and steric effects in the model. We extend the classical Scharfetter-Gummel (SG) method for semiconductor devices to include the steric potential for ion channels, which is a fundamental physical property not present in semiconductors. Together with a simplified matched interface and boundary (SMIB) method for treating molecular surfaces and singular charges of channel proteins, the extended SG method is shown to exhibit important features in flow simulations such as optimal convergence, efficient nonlinear iterations, and physical conservation. Two different methods --- called the SMIB and multiscale methods --- are proposed for two different types of channels, namely, the gramicidin A channel and an L-type calcium channel, depending on whether water is allowed to pass through the channel. The PNPF currents are in accord with the experimental I-V (V for applied voltages) data of the gramicidin A channel and I-C (C for bath concentrations) data of the calcium channel.

q-bio.BM↗