arXiv subjects

Tao Du

Publications and source records attributed to Tao Du.

47 records · Page 3Linked to original sources

arXiv2021

DiffAqua: A Differentiable Computational Design Pipeline for Soft Underwater Swimmers with Shape Interpolation

The computational design of soft underwater swimmers is challenging because of the high degrees of freedom in soft-body modeling. In this paper, we present a differentiable pipeline for co-designing a soft swimmer's geometry and controller. Our pipeline unlocks gradient-based algorithms for discovering novel swimmer designs more efficiently than traditional gradient-free solutions. We propose Wasserstein barycenters as a basis for the geometric design of soft underwater swimmers since it is differentiable and can naturally interpolate between bio-inspired base shapes via optimal transport. By combining this design space with differentiable simulation and control, we can efficiently optimize a soft underwater swimmer's performance with fewer simulations than baseline methods. We demonstrate the efficacy of our method on various design problems such as fast, stable, and energy-efficient swimming and demonstrate applicability to multi-objective design.

cs.LG↗

arXiv2021

PlasticineLab: A Soft-Body Manipulation Benchmark with Differentiable Physics

Simulated virtual environments serve as one of the main driving forces behind developing and evaluating skill learning algorithms. However, existing environments typically only simulate rigid body physics. Additionally, the simulation process usually does not provide gradients that might be useful for planning and control optimizations. We introduce a new differentiable physics benchmark called PasticineLab, which includes a diverse collection of soft body manipulation tasks. In each task, the agent uses manipulators to deform the plasticine into the desired configuration. The underlying physics engine supports differentiable elastic and plastic deformation using the DiffTaichi system, posing many under-explored challenges to robotic agents. We evaluate several existing reinforcement learning (RL) methods and gradient-based methods on this benchmark. Experimental results suggest that 1) RL-based approaches struggle to solve most of the tasks efficiently; 2) gradient-based approaches, by optimizing open-loop control sequences with the built-in differentiable physics engine, can rapidly find a solution within tens of iterations, but still fall short on multi-stage tasks that require long-term planning. We expect that PlasticineLab will encourage the development of novel algorithms that combine differentiable physics and RL for more complex physics-based skill learning tasks.

cs.LG↗

arXiv2021

Effects of correlations on phase diagrams of the two-dimensional Su-Schrieffer-Heeger model with the longer-range hoppings

For the Su-Schrieffer-Heeger (SSH) model on the two-dimensional square lattice, two third nearest neighbor hoppings which preserve chiral symmetry are introduced. Like the case of one dimension, the longer-range hopping can drive topological transitions and leads to the larger topological invariant (vectored Berry phase for the two dimensional SSH model). We obtain the phase boundaries and topological phase diagrams from the winding pattern of $ρ(k_l)$ of the Bloch Hamiltonian. Effects of correlations on the extended two dimensional SSH model are also investigated. The correlation shifts the phase boundaries and leads to topological transitions. Several special points in non-interacting phase diagrams are chosen to illustrate the different phase transitions and a correlations-driven topologically non-trivial phase from the trivial one is found. In this work, the slave-rotor mean field method is applied to the interacting model and we focus on the case of physical electrons, i.e. the Mott transition of charge degree of freedom is avoided in our investigations.

cond-mat.str-el↗

arXiv2021

Phase transitions of the dimerized Kane-Mele model with/without the strong interaction

The dimerized Kane-Mele model with/without the strong interaction is studied using analytical methods. The boundary of the topological phase transition of the model without strong interaction is obtained. Our results show that the occurrence of the transition only depends on dimerized parameter . From the one-particle spectrum, we obtain the completed phase diagram including the quantum spin Hall (QSH) state and the topologically trivial insulator. Then, using different mean-field methods, we investigate the Mott transition and the magnetic transition of the strongly correlated dimerized Kane-Mele model. In the region between the two transitions, the topological Mott insulator (TMI) with characters of Mott insulators and topological phases may be the most interesting phase. In this work, effects of the hopping anisotropy and Hubbard interaction U on boundaries of the two transitions are observed in detail. The completed phase diagram of the dimerized Kane-Mele-Hubbard model is also obtained in this work. Quantum fluctuations have extremely important influences on a quantum system. However, investigations are under the framework of the mean field treatment in this work and the effects of fluctuations in this model will be discussed in the future.

cond-mat.str-el↗

arXiv2021

The competition between the intrinsic and Rashba spin-orbit coupling and effects of correlations on Rashba SOC-driven transitions in the Kane-Mele model

We investigate, firstly, the effects of the Rashba SOC on the band structrue of the Kane-Mele model. The competition between the Rashba SOC and the intrinsic SOC can lead to the rich phenomenology. The Rashba SOC can drive the indirect and direct energy gap to close successively, but maintain the band touching between the valence band and the conduction band when the Rashba SOC is large enough to dominant the competition. We find that these touching points are located at $K$ and $K^{\prime}$ or/and some $2π/3$ rotationally symmetric points around $K$ and $K^{\prime}$ in the Brillouin zone. The indirect and direct energy gap closings correspond to the topologically trivial and non-trivial phase transitions respectively. For the small intrinsic SOC, the topologically non-trivial transition occurs when the ratio of the Rashba SOC to the intrinsic SOC is equal to the classical result, i.e. $2\sqrt{3}$. For the large intrinsic SOC, however, we find that the ratio decreases with the increasing intrinsic SOC. Secondly, using the slave-rotor mean field method we investigate the influences of the correlation on the Rashba SOC-driven topologically trivial and non-trivial transition in both the charge condensate and Mott regions. The topological Mott insulator with gapped or gapless spin excitations can arise from the interplay of the Rashba SOC and correlations.

cond-mat.str-el↗

arXiv2020

Efficient Continuous Pareto Exploration in Multi-Task Learning

Tasks in multi-task learning often correlate, conflict, or even compete with each other. As a result, a single solution that is optimal for all tasks rarely exists. Recent papers introduced the concept of Pareto optimality to this field and directly cast multi-task learning as multi-objective optimization problems, but solutions returned by existing methods are typically finite, sparse, and discrete. We present a novel, efficient method that generates locally continuous Pareto sets and Pareto fronts, which opens up the possibility of continuous analysis of Pareto optimal solutions in machine learning problems. We scale up theoretical results in multi-objective optimization to modern machine learning problems by proposing a sample-based sparse linear system, for which standard Hessian-free solvers in machine learning can be applied. We compare our method to the state-of-the-art algorithms and demonstrate its usage of analyzing local Pareto sets on various multi-task classification and regression problems. The experimental results confirm that our algorithm reveals the primary directions in local Pareto sets for trade-off balancing, finds more solutions with different trade-offs efficiently, and scales well to tasks with millions of parameters.

cs.LG↗

arXiv2020

On the Allowable or Forbidden Nature of Vapor-Deposited Glasses

Vapor deposition can yield glasses that are more stable than those obtained by the traditional melt-quenching route. However, it remains unclear whether vapor-deposited glasses are "allowable" or "forbidden," that is, if they are equivalent to glasses formed by cooling extremely slowly a liquid or if they differ in nature from melt-quenched glasses. Here, based on reactive molecular dynamics simulation (MD) of silica glasses, we demonstrate that the allowable or forbidden nature of vapor-deposited glasses depends on the temperature of the substrate and, in turn, is found to be encoded in their medium-range order structure.

cond-mat.mtrl-sci↗

arXiv2019

Self-driving scale car trained by Deep reinforcement learning

The self-driving based on deep reinforcement learning, as the most important application of artificial intelligence, has become a popular topic. Most of the current self-driving methods focus on how to directly learn end-to-end self-driving control strategy from the raw sensory data. Essentially, this control strategy can be considered as a mapping between images and driving behavior, which usually faces a problem of low generalization ability. To improve the generalization ability for the driving behavior, the reinforcement learning method requires extrinsic reward from the real environment, which may damage the car. In order to obtain a good generalization ability in safety, a virtual simulation environment that can be constructed different driving scene is designed by Unity. A theoretical model is established and analyzed in the virtual simulation environment, and it is trained by double Deep Q-network. Then, the trained model is migrated to a scale car in real world. This process is also called a sim2real method. The sim2real training method efficiently handle the these two problems. The simulations and experiments are carried out to evaluate the performance and effectiveness of the proposed algorithm. Finally, it is demonstrated that the scale car in real world obtain the capability for autonomous driving.

cs.LG↗

arXiv2019

Revealing the Atomic Structure of Silicate Glasses by Force-Enhanced Atomic Refinement

Although experiments can offer some fingerprints of the atomic structure of glasses (coordination numbers, pair distribution function, etc.), atomistic simulations are often required to directly access the structure itself (i.e., the positions of the atoms). On the one hand, molecular dynamics (MD) simulations can be used to generate by quenching a liquid - but MD simulations remain plagued by extremely high cooling rates. On the other hand, reverse Monte Carlo (RMC) modeling bypasses the melt-quenching route - but RMC often yields non-unique glass structures. Here, we adopt the force-enhanced atomic refinement (FEAR) method to overcome these limitations and decipher the atomic structure of a sodium silicate glass. We show that FEAR offers an unprecedented description of the atomic structure of sodium silicate. The FEAR-generated glass structure simultaneously exhibits (i) enhanced agreement with experimental neutron diffraction data and (ii) higher energetic stability as compared to those generated by MD or RMC. This result allows us to reveal new insights into the atomic structure of sodium silicate glasses. Specifically, we show that sodium silicate glasses exhibit a more ordered medium-range order structure than previously suggested by MD simulations. These results pave the way toward an increased ability to accurately describe the atomic structure of glasses.

cond-mat.mtrl-sci↗

arXiv2018

Generation of mode-2 internal waves in a two-dimensional stratification by a mode-1 internal wave

The generation of mode-2 nonlinear internal waves (IWs) by the evolution of a mode-1 IW in a two-dimensional stratification is investigated. A generation model accounting for intermodal interaction is derived based on a multi-modal approach in a weakly nonlinear and non-hydrostatic configuration. The generation model is numerically solved to simulate the evolution of mode-1 and mode-2 IWs in an inhomogeneous pycnocline. The numerical experiments confirm that a mode-2 IW is generated due to linear and nonlinear intermodal interaction. The mode-2 IW continues growing and gradually separates with the mode-1 IW during the generation process. A non-dimensional quantity quantifying the energy of mode-2 IWs is used to investigate the favorable conditions for the formation of mode-2 IWs. The numerical results suggest that the pycnocline strength or depth prominently affects the formation of mode-2 IWs, followed by pycnocline thickness. A weakening or shoaling pycnocline favors the formation of mode-2 IWs by evidently enhancing linear and nonlinear intermodal interaction, whereas a thinning pycnocline favors the process mainly by enhancing nonlinear intermodal interaction. Shortening the front length inhibits nonlinear intermodal interaction while equivalently strengthens the linear intermodal interaction. Increasing the initial mode-1 IW amplitude can noticeably increase the produced mode-2 IW amplitude.

physics.ao-ph↗

arXiv2017

Multitude of phases in correlated lattice fermion systems with spin-dependent disorder

The magnetic phases induced by the interplay between disorder acting only on particles with a given spin projection ("spin-dependent disorder") and a local repulsive interaction is explored. To this end the magnetic ground state phase diagram of the Hubbard model at half-filling is computed within dynamical mean-field theory combined with the geometric average over disorder, which is able to describe Anderson localization. Five distinct phases are identified: a ferromagnetically polarized metal, two types of insulators, and two types of spin-selective localized phases. The latter four phases possess different long-range order of the spins. The predicted phase diagram may be tested experimentally using cold fermions in optical lattices subject to spin-dependent random potentials.

cond-mat.str-el↗