arXiv subjects

Bowen Liu

Publications and source records attributed to Bowen Liu.

118 records · Page 7Linked to original sources

arXiv2020

Weakly-Supervised Salient Object Detection via Scribble Annotations

Compared with laborious pixel-wise dense labeling, it is much easier to label data by scribbles, which only costs 1$\sim$2 seconds to label one image. However, using scribble labels to learn salient object detection has not been explored. In this paper, we propose a weakly-supervised salient object detection model to learn saliency from such annotations. In doing so, we first relabel an existing large-scale salient object detection dataset with scribbles, namely S-DUTS dataset. Since object structure and detail information is not identified by scribbles, directly training with scribble labels will lead to saliency maps of poor boundary localization. To mitigate this problem, we propose an auxiliary edge detection task to localize object edges explicitly, and a gated structure-aware loss to place constraints on the scope of structure to be recovered. Moreover, we design a scribble boosting scheme to iteratively consolidate our scribble annotations, which are then employed as supervision to learn high-quality saliency maps. As existing saliency evaluation metrics neglect to measure structure alignment of the predictions, the saliency map ranking metric may not comply with human perception. We present a new metric, termed saliency structure measure, to measure the structure alignment of the predicted saliency maps, which is more consistent with human perception. Extensive experiments on six benchmark datasets demonstrate that our method not only outperforms existing weakly-supervised/unsupervised methods, but also is on par with several fully-supervised state-of-the-art models. Our code and data is publicly available at https://github.com/JingZhang617/Scribble_Saliency.

cs.CV↗

arXiv2020

Fail-safe Watchtowers and Short-lived Assertions for Payment Channels

The recent development of payment channels and their extensions (e.g., state channels) provides a promising scalability solution for blockchains which allows untrusting parties to transact off-chain and resolve potential disputes via on-chain smart contracts. To protect participants who have no constant access to the blockchain, a watching service named as watchtower is proposed -- a third-party entity obligated to monitor channel states (on behalf of the participants) and correct them on-chain if necessary. Unfortunately, currently proposed watchtower schemes suffer from multiple security and efficiency drawbacks. In this paper, we explore the design space behind watchtowers. We propose a novel watching service named as fail-safe watchtowers. In contrast to prior proposed watching services, our fail-safe watchtower does not watch on-chain smart contracts constantly. Instead, it only sends a single on-chain message periodically confirming or denying the final states of channels being closed. Our watchtowers can easily handle a large number of channels, are privacy-preserving, and fail-safe tolerating multiple attack vectors. Furthermore, we show that watchtowers (in general) may be an option economically unjustified for multiple payment scenarios and we introduce a simple, yet powerful concept of short-lived assertions which can mitigate misbehaving parties in these scenarios.

cs.CR↗

arXiv2020

Strategies for Pre-training Graph Neural Networks

Many applications of machine learning require a model to make accurate pre-dictions on test examples that are distributionally different from training ones, while task-specific labels are scarce during training. An effective approach to this challenge is to pre-train a model on related tasks where data is abundant, and then fine-tune it on a downstream task of interest. While pre-training has been effective in many language and vision domains, it remains an open question how to effectively use pre-training on graph datasets. In this paper, we develop a new strategy and self-supervised methods for pre-training Graph Neural Networks (GNNs). The key to the success of our strategy is to pre-train an expressive GNN at the level of individual nodes as well as entire graphs so that the GNN can learn useful local and global representations simultaneously. We systematically study pre-training on multiple graph classification datasets. We find that naive strategies, which pre-train GNNs at the level of either entire graphs or individual nodes, give limited improvement and can even lead to negative transfer on many downstream tasks. In contrast, our strategy avoids negative transfer and improves generalization significantly across downstream tasks, leading up to 9.4% absolute improvements in ROC-AUC over non-pre-trained models and achieving state-of-the-art performance for molecular property prediction and protein function prediction.

cs.LG↗

arXiv2019

Unified Signal Compression Using Generative Adversarial Networks

We propose a unified compression framework that uses generative adversarial networks (GAN) to compress image and speech signals. The compressed signal is represented by a latent vector fed into a generator network which is trained to produce high quality signals that minimize a target objective function. To efficiently quantize the compressed signal, non-uniformly quantized optimal latent vectors are identified by iterative back-propagation with ADMM optimization performed for each iteration. Our experiments show that the proposed algorithm outperforms prior signal compression methods for both image and speech compression quantified in various metrics including bit rate, PSNR, and neural network based signal classification accuracy.

eess.SP↗

arXiv2019

Linear Stability of Elliptic Relative Equilibria of Restricted Four-body Problem

In this paper, we consider the linear stability of the elliptic relative equilibria of the restricted 4-body problems where the three primaries form a Lagrangian triangle. By reduction, the linearized Poincaré map is decomposed to the essential part, the Keplerian part and the elliptic Lagrangian part where the last two parts have been studied in literature. The linear stability of the essential part depends on the masses parameters $α$, $β$ with $α\geq β>0$ and the eccentricity $e\in[0,1)$. Via $\om$-Maslov index theory and linear differential operator theory, we obtain the full bifurcation diagram of linearly stable and unstable regions with respect to $α$, $β$ and $e$. Especially, two linearly stable sub-regions are found.

math-ph↗

arXiv2019

Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation

Generating novel graph structures that optimize given objectives while obeying some given underlying rules is fundamental for chemistry, biology and social science research. This is especially important in the task of molecular graph generation, whose goal is to discover novel molecules with desired properties such as drug-likeness and synthetic accessibility, while obeying physical laws such as chemical valency. However, designing models to find molecules that optimize desired properties while incorporating highly complex and non-differentiable rules remains to be a challenging task. Here we propose Graph Convolutional Policy Network (GCPN), a general graph convolutional network based model for goal-directed graph generation through reinforcement learning. The model is trained to optimize domain-specific rewards and adversarial loss through policy gradient, and acts in an environment that incorporates domain-specific rules. Experimental results show that GCPN can achieve 61% improvement on chemical property optimization over state-of-the-art baselines while resembling known molecules, and achieve 184% improvement on the constrained property optimization task.

cs.LG↗

arXiv2019

Band Structure Engineering of Interfacial Semiconductors Based on Atomically Thin Lead Iodide Crystals

To explore new constituents in two-dimensional materials and to combine their best in van der Waals heterostructures, are in great demand as being unique platform to discover new physical phenomena and to design novel functionalities in interface-based devices. Herein, PbI2 crystals as thin as few-layers are first synthesized, particularly through a facile low-temperature solution approach with the crystals of large size, regular shape, different thicknesses and high-yields. As a prototypical demonstration of flexible band engineering of PbI2-based interfacial semiconductors, these PbI2 crystals are subsequently assembled with several transition metal dichalcogenide monolayers. The photoluminescence of MoS2 is strongly enhanced in MoS2/PbI2 stacks, while a dramatic photoluminescence quenching of WS2 and WSe2 is revealed in WS2/PbI2 and WSe2/PbI2 stacks. This is attributed to the effective heterojunction formation between PbI2 and these monolayers, but type I band alignment in MoS2/PbI2 stacks where fast-transferred charge carriers accumulate in MoS2 with high emission efficiency, and type II in WS2/PbI2 and WSe2/PbI2 stacks with separated electrons and holes suitable for light harvesting. Our results demonstrate that MoS2, WS2, WSe2 monolayers with very similar electronic structures themselves, show completely distinct light-matter interactions when interfacing with PbI2, providing unprecedent capabilities to engineer the device performance of two-dimensional heterostructures.

cond-mat.mes-hall↗

arXiv2018

Homographic Solutions of the N-body Generalized Lennard-Jones System

In this paper, we obtain the existence of non-planar circular homographic solutions and non-circular homographic solutions of the $(2+N)$- and $(3+N)$-body problems of the Lennard-Jones system. These results show the essential difference between the Lennard-Jones potential and the Newton's potential of universal gravitation.

math.DS↗

arXiv2017

Retrosynthetic reaction prediction using neural sequence-to-sequence models

We describe a fully data driven model that learns to perform a retrosynthetic reaction prediction task, which is treated as a sequence-to-sequence mapping problem. The end-to-end trained model has an encoder-decoder architecture that consists of two recurrent neural networks, which has previously shown great success in solving other sequence-to-sequence prediction tasks such as machine translation. The model is trained on 50,000 experimental reaction examples from the United States patent literature, which span 10 broad reaction types that are commonly used by medicinal chemists. We find that our model performs comparably with a rule-based expert system baseline model, and also overcomes certain limitations associated with rule-based expert systems and with any machine learning approach that contains a rule-based expert system component. Our model provides an important first step towards solving the challenging problem of computational retrosynthetic analysis.

cs.LG↗

arXiv2017

Group-velocity-locked vector soliton molecules in a birefringence-enhanced fiber laser

Physics phenomena of multi-soliton complexes have enriched the life of dissipative solitons in fiber lasers. By developing a birefringence-enhanced fiber laser, we report the first experimental observation of group-velocity-locked vector soliton (GVLVS) molecules. The birefringence-enhanced fiber laser facilitates the generation of GVLVSs, where the two orthogonally polarized components are coupled together to form a multi-soliton complex. Moreover, the interaction of repulsive and attractive forces between multiple pulses binds the particle-like GVLVSs together in time domain to further form compound multi-soliton complexes, namely GVLVS molecules. By adopting the polarization-resolved measurement, we show that the two orthogonally polarized components of the GVLVS molecules are both soliton molecules supported by the strongly modulated spectral fringes and the double-humped intensity profiles. Additionally, GVLVS molecules with various soliton separations are also observed by adjusting the pump power and the polarization controller.

physics.optics↗