arXiv subjects
Publications and source records attributed to Zhen Li.
In this paper, we formulate the deep residual network (ResNet) as a control problem of transport equation. In ResNet, the transport equation is solved along the characteristics. Based on this observation, deep neural network is closely related to the control problem of PDEs on manifold. We propose several models based on transport equation and Hamilton-Jacobi equation. The discretization of these PDEs on point cloud is also discussed.
The automatic detection of software vulnerabilities is an important research problem. However, existing solutions to this problem rely on human experts to define features and often miss many vulnerabilities (i.e., incurring high false negative rate). In this paper, we initiate the study of using deep learning-based vulnerability detection to relieve human experts from the tedious and subjective task of manually defining features. Since deep learning is motivated to deal with problems that are very different from the problem of vulnerability detection, we need some guiding principles for applying deep learning to vulnerability detection. In particular, we need to find representations of software programs that are suitable for deep learning. For this purpose, we propose using code gadgets to represent programs and then transform them into vectors, where a code gadget is a number of (not necessarily consecutive) lines of code that are semantically related to each other. This leads to the design and implementation of a deep learning-based vulnerability detection system, called Vulnerability Deep Pecker (VulDeePecker). In order to evaluate VulDeePecker, we present the first vulnerability dataset for deep learning approaches. Experimental results show that VulDeePecker can achieve much fewer false negatives (with reasonable false positives) than other approaches. We further apply VulDeePecker to 3 software products (namely Xen, Seamonkey, and Libav) and detect 4 vulnerabilities, which are not reported in the National Vulnerability Database but were "silently" patched by the vendors when releasing later versions of these products; in contrast, these vulnerabilities are almost entirely missed by the other vulnerability detection systems we experimented with.
Based on a natural connection between ResNet and transport equation or its characteristic equation, we propose a continuous flow model for both ResNet and plain net. Through this continuous model, a ResNet can be explicitly constructed as a refinement of a plain net. The flow model provides an alternative perspective to understand phenomena in deep neural networks, such as why it is necessary and sufficient to use 2-layer blocks in ResNets, why deeper is better, and why ResNets are even deeper, and so on. It also opens a gate to bring in more tools from the huge area of differential equations.
Fine-grained image labels are desirable for many computer vision applications, such as visual search or mobile AI assistant. These applications rely on image classification models that can produce hundreds of thousands (e.g. 100K) of diversified fine-grained image labels on input images. However, training a network at this vocabulary scale is challenging, and suffers from intolerable large model size and slow training speed, which leads to unsatisfying classification performance. A straightforward solution would be training separate expert networks (specialists), with each specialist focusing on learning one specific vertical (e.g. cars, birds...). However, deploying dozens of expert networks in a practical system would significantly increase system complexity and inference latency, and consumes large amounts of computational resources. To address these challenges, we propose a Knowledge Concentration method, which effectively transfers the knowledge from dozens of specialists (multiple teacher networks) into one single model (one student network) to classify 100K object categories. There are three salient aspects in our method: (1) a multi-teacher single-student knowledge distillation framework; (2) a self-paced learning mechanism to allow the student to learn from different teachers at various paces; (3) structurally connected layers to expand the student network capacity with limited extra parameters. We validate our method on OpenImage and a newly collected dataset, Entity-Foto-Tree (EFT), with 100K categories, and show that the proposed model performs significantly better than the baseline generalist model.
Recurrent neural networks (RNNs) have been successfully applied to various natural language processing (NLP) tasks and achieved better results than conventional methods. However, the lack of understanding of the mechanisms behind their effectiveness limits further improvements on their architectures. In this paper, we present a visual analytics method for understanding and comparing RNN models for NLP tasks. We propose a technique to explain the function of individual hidden state units based on their expected response to input texts. We then co-cluster hidden state units and words based on the expected response and visualize co-clustering results as memory chips and word clouds to provide more structured knowledge on RNNs' hidden states. We also propose a glyph-based sequence visualization based on aggregate information to analyze the behavior of an RNN's hidden state at the sentence-level. The usability and effectiveness of our method are demonstrated through case studies and reviews from domain experts.
We propose a data-driven method for recovering miss-ing parts of 3D shapes. Our method is based on a new deep learning architecture consisting of two sub-networks: a global structure inference network and a local geometry refinement network. The global structure inference network incorporates a long short-term memorized context fusion module (LSTM-CF) that infers the global structure of the shape based on multi-view depth information provided as part of the input. It also includes a 3D fully convolutional (3DFCN) module that further enriches the global structure representation according to volumetric information in the input. Under the guidance of the global structure network, the local geometry refinement network takes as input lo-cal 3D patches around missing regions, and progressively produces a high-resolution, complete surface through a volumetric encoder-decoder architecture. Our method jointly trains the global structure inference and local geometry refinement networks in an end-to-end manner. We perform qualitative and quantitative evaluations on six object categories, demonstrating that our method outperforms existing state-of-the-art work on shape completion.
We simulate complex fluids by means of an on-the-fly coupling of the bulk rheology to the underlying microstructure dynamics. In particular, a macroscopic continuum model of polymeric fluids is constructed without a pre-specified constitutive relation, but instead it is actively learned from mesoscopic simulations where the dynamics of polymer chains is explicitly computed. To couple the macroscopic rheology of polymeric fluids and the microscale dynamics of polymer chains, the continuum approach (based on the finite volume method) provides the transient flow field as inputs for the (mesoscopic) dissipative particle dynamics (DPD), and in turn DPD returns an effective constitutive relation to close the continuum equations. In this multiscale modeling procedure, we employ an active learning strategy based on Gaussian process regression (GPR) to minimize the number of expensive DPD simulations, where adaptively selected DPD simulations are performed only as necessary. Numerical experiments are carried out for flow past a circular cylinder of a non-Newtonian fluid, modeled at the mesoscopic level by bead-spring chains. The results show that only five DPD simulations are required to achieve an effective closure of the continuum equations at Reynolds number Re=10. Furthermore, when Re is increased to 100, only one additional DPD simulation is required for constructing an extended GPR-informed model closure. Compared to traditional message-passing multiscale approaches, applying an active learning scheme to multiscale modeling of non-Newtonian fluids can significantly increase the computational efficiency. Although the method demonstrated here obtains only a local viscosity from the mesoscopic model, it can be extended to other multiscale models of complex fluids whose macro-rheology is unknown.
Computational elucidation of membrane protein (MP) structures is challenging partially due to lack of sufficient solved structures for homology modeling. Here we describe a high-throughput deep transfer learning method that first predicts MP contacts by learning from non-membrane proteins (non-MPs) and then predicting three-dimensional structure models using the predicted contacts as distance restraints. Tested on 510 non-redundant MPs, our method has contact prediction accuracy at least 0.18 better than existing methods, predicts correct folds for 218 MPs (TMscore at least 0.6), and generates three-dimensional models with RMSD less than 4 Angstrom and 5 Angstrom for 57 and 108 MPs, respectively. A rigorous blind test in the continuous automated model evaluation (CAMEO) project shows that our method predicted high-resolution three-dimensional models for two recent test MPs of 210 residues with RMSD close to 2 Angstrom. We estimated that our method could predict correct folds for between 1,345 and 1,871 reviewed human multi-pass MPs including a few hundred new folds, which shall facilitate the discovery of drugs targeting at membrane proteins.
While pyrochlore iridate thin films are theoretically predicted to possess a variety of emergent topological properties, experimental verification of these predictions can be obstructed by the challenge in thin film growth. Here we report on the pulsed laser deposition and characterization of thin films of a representative pyrochlore compound Bi2Ir2O7. The films were epitaxially grown on yttria-stabilized zirconia substrates and have lattice constants that are a few percent larger than that of the bulk single crystals. The film composition shows a strong dependence on the oxygen partial pressure. Density-functional-theory calculations indicate the existence of Bi_Ir antisite defects, qualitatively consistent with the high Bi: Ir ratio found in the films. Both Ir and Bi have oxidation states that are lower than their nominal values, suggesting the existence of oxygen deficiency. The iridate thin films show a variety of intriguing transport characteristics, including multiple charge carriers, logarithmic dependence of resistance on temperature, antilocalization corrections to conductance due to spin-orbit interactions, and linear positive magnetoresistance.
Optomechanical quantum interfaces can be utilized to connect systems with distinctively different frequencies in hybrid quantum networks. Here we present a scheme of nonreciprocal quantum state conversion between microwave and optical photons via an optomechanical interface. By introducing an auxiliary cavity and manipulating the phase differences between the linearized optomechanical couplings, uni-directional state transmission can be achieved. The interface can function as an isolator, a circulator, and a two-way switch that routes the input states to a selected output channel. We show that under a generalized impedance matching condition, the state conversion can reach high fidelity and is robust against the thermal fluctuations in the mechanical mode.
We develop a new method called Discriminated Hub Graphical Lasso (DHGL) based on Hub Graphical Lasso (HGL) by providing prior information of hubs. We apply this new method in two situations: with known hubs and without known hubs. Then we compare DHGL with HGL using several measures of performance. When some hubs are known, we can always estimate the precision matrix better via DHGL than HGL. When no hubs are known, we use Graphical Lasso (GL) to provide information of hubs and find that the performance of DHGL will always be better than HGL if correct prior information is given and will seldom degenerate when the prior information is wrong.
Computational prediction of membrane protein (MP) structures is very challenging partially due to lack of sufficient solved structures for homology modeling. Recently direct evolutionary coupling analysis (DCA) sheds some light on protein contact prediction and accordingly, contact-assisted folding, but DCA is effective only on some very large-sized families since it uses information only in a single protein family. This paper presents a deep transfer learning method that can significantly improve MP contact prediction by learning contact patterns and complex sequence-contact relationship from thousands of non-membrane proteins (non-MPs). Tested on 510 non-redundant MPs, our deep model (learned from only non-MPs) has top L/10 long-range contact prediction accuracy 0.69, better than our deep model trained by only MPs (0.63) and much better than a representative DCA method CCMpred (0.47) and the CASP11 winner MetaPSICOV (0.55). The accuracy of our deep model can be further improved to 0.72 when trained by a mix of non-MPs and MPs. When only contacts in transmembrane regions are evaluated, our method has top L/10 long-range accuracy 0.62, 0.57, and 0.53 when trained by a mix of non-MPs and MPs, by non-MPs only, and by MPs only, respectively, still much better than MetaPSICOV (0.45) and CCMpred (0.40). All these results suggest that sequence-structure relationship learned by our deep model from non-MPs generalizes well to MP contact prediction. Improved contact prediction also leads to better contact-assisted folding. Using only top predicted contacts as restraints, our deep learning method can fold 160 and 200 of 510 MPs with TMscore>0.6 when trained by non-MPs only and by a mix of non-MPs and MPs, respectively, while CCMpred and MetaPSICOV can do so for only 56 and 77 MPs, respectively. Our contact-assisted folding also greatly outperforms homology modeling.
We report a terahertz optical modulator consisting of randomly stacked trilayer graphene (TLG) deposited on an oxidized silicon substrate by means of THz- Time Domain Spectroscopy (THz-TDS). Here, the gate tuning of the Fermi level of the TLG provides the fundamental basis for the modulation of THz transmission. We measured a 15% change in the THz transmission of this device over a broad frequency range (0.6-1.6THz). We also observed a strong absorption >80% in the time-domain signals and a frequency independence of the conductivity. Furthermore, unlike previous studies, we find that the underlying silicon substrate, which serves as a gate electrode for the graphene, also exhibits substantial modulation of the transmitted THz radiation under applied voltage biases.
We present a local detection method for dissipative particle dynamics (DPD) involving arbitrarily shaped geometric three-dimensional domains. By introducing an indicator variable of boundary volume fraction (BVF) for each fluid particle, the boundary of arbitrary-shape objects is detected on-the-fly for the moving fluid particles using only the local particle configuration. Therefore, this approach eliminates the need of an analytical description of the boundary and geometry of objects in DPD simulations and makes it possible to load the geometry of a system directly from experimental images or computer-aided designs/drawings. Wall penetration is inferred from the value of the BVF and prevented by a predictor-corrector algorithm. The no-slip boundary condition is achieved by employing effective dissipative coefficients for liquid-solid interactions. Quantitative evaluations of the new method are performed for the plane Poiseuille flow, the plane Couette flow and the Wannier flow in a cylindrical domain and compared with their corresponding analytical solutions and (high-order) spectral element solution of the Navier-Stokes equations. We verify that the proposed method yields correct no-slip boundary condition for velocity and generates negligible fluctuations of density and temperature in the vicinity of the wall surface. Moreover, we construct a very complex 3D geometry - the "Brown Pacman" microfluidic device - to explicitly demonstrate how to construct a DPD system with complex geometry directly from loading a graphical image. In addition to stationary arbitrary-shape objects, the new method is particularly useful for problems involving moving and deformable boundaries, because it only uses local information of neighboring particles and satisfies the desired boundary conditions on-the-fly.
Oxygen is widely used to tune the performance of chalcogenide phase-change materials in the usage of phase-Change random access memory (PCRAM) which is considered as the most promising next-generation non-volatile memory. However, the microscopic role of oxygen in the write-erase process, i.e., the reversible phase transition between crystalline and amorphous state of phase-change materials is not clear yet. Using oxygen doped GeTe as an example, this work unravels the role of oxygen at the atomic scale by means of ab initio total energy calculations and ab initio molecular dynamics simulations. Our main finding is that after the amorphization and the subsequent re-crystallization process simulated by ab initio molecular dynamics, oxygen will drag one Ge atom out of its lattice site and both atoms stay in the interstitial region near the Te vacancy that was originally occupied by the oxygen, forming a "dumbbell-like" defect (O-VTe-Ge), which is in sharp contrast to the results of ab initio total energy calculations at 0 K showing that the oxygen prefers to substitute Te in crystalline GeTe. This specific defect configuration is found to be responsible for the slower crystallization speed and hence the improved data retention of oxygen doped GeTe as reported in recent experimental work. Moreover, we find that the oxygen will increase the effective mass of the carrier and thus increases the resistivity of GeTe. Our results unravel the microscopic mechanism of the oxygen-doping optimization of phase-change material GeTe, and the present reported mechanism can be applied to other oxygen doped ternary chalcogenide phase-change materials.
Recently exciting progress has been made on protein contact prediction, but the predicted contacts for proteins without many sequence homologs is still of low quality and not very useful for de novo structure prediction. This paper presents a new deep learning method that predicts contacts by integrating both evolutionary coupling (EC) and sequence conservation information through an ultra-deep neural network formed by two deep residual networks. This deep neural network allows us to model very complex sequence-contact relationship as well as long-range inter-contact correlation. Our method greatly outperforms existing contact prediction methods and leads to much more accurate contact-assisted protein folding. Tested on three datasets of 579 proteins, the average top L long-range prediction accuracy obtained our method, the representative EC method CCMpred and the CASP11 winner MetaPSICOV is 0.47, 0.21 and 0.30, respectively; the average top L/10 long-range accuracy of our method, CCMpred and MetaPSICOV is 0.77, 0.47 and 0.59, respectively. Ab initio folding using our predicted contacts as restraints can yield correct folds (i.e., TMscore>0.6) for 203 test proteins, while that using MetaPSICOV- and CCMpred-predicted contacts can do so for only 79 and 62 proteins, respectively. Further, our contact-assisted models have much better quality than template-based models. Using our predicted contacts as restraints, we can (ab initio) fold 208 of the 398 membrane proteins with TMscore>0.5. By contrast, when the training proteins of our method are used as templates, homology modeling can only do so for 10 of them. One interesting finding is that even if we do not train our prediction models with any membrane proteins, our method works very well on membrane protein prediction. Finally, in recent blind CAMEO benchmark our method successfully folded 5 test proteins with a novel fold.
Mesoscopic numerical simulations provide a unique approach for the quantification of the chemical influences on red blood cell functionalities. The transport Dissipative Particles Dynamics (tDPD) method can lead to such effective multiscale simulations due to its ability to simultaneously capture mesoscopic advection, diffusion, and reaction. In this paper, we present a GPU-accelerated red blood cell simulation package based on a tDPD adaptation of our red blood cell model, which can correctly recover the cell membrane viscosity, elasticity, bending stiffness, and cross-membrane chemical transport. The package essentially processes all computational workloads in parallel by GPU, and it incorporates multi-stream scheduling and non-blocking MPI communications to improve inter-node scalability. Our code is validated for accuracy and compared against the CPU counterpart for speed. Strong scaling and weak scaling are also presented to characterizes scalability. We observe a speedup of 10.1 on one GPU over all 16 cores within a single node, and a weak scaling efficiency of 91% across 256 nodes. The program enables quick-turnaround and high-throughput numerical simulations for investigating chemical-driven red blood cell phenomena and disorders.
There are many orbital excited mesons discovered in recent years. In this work we attempt to study whether the Regge trajectory is quasi-linear or square-root form. In the framework of the quasi-linear Regge trajectory and square-root Regge trajectory, the masses of the states lying on the well established 11S0, 13S1, and 13P2 trajectories are estimated. Comparison of the results given by the two trajectories with the existing experimental data illustrates that both of them can give a reasonable description for the ground mesons. For the orbital excited states, the quasi-linear trajectory describes the existing meson spectrum to be more reasonable.