arXiv subjects

Jian Jiang

Publications and source records attributed to Jian Jiang.

42 records · Page 3Linked to original sources

arXiv2021

DeepFaceLab: Integrated, flexible and extensible face-swapping framework

Deepfake defense not only requires the research of detection but also requires the efforts of generation methods. However, current deepfake methods suffer the effects of obscure workflow and poor performance. To solve this problem, we present DeepFaceLab, the current dominant deepfake framework for face-swapping. It provides the necessary tools as well as an easy-to-use way to conduct high-quality face-swapping. It also offers a flexible and loose coupling structure for people who need to strengthen their pipeline with other features without writing complicated boilerplate code. We detail the principles that drive the implementation of DeepFaceLab and introduce its pipeline, through which every aspect of the pipeline can be modified painlessly by users to achieve their customization purpose. It is noteworthy that DeepFaceLab could achieve cinema-quality results with high fidelity. We demonstrate the advantage of our system by comparing our approach with other face-swapping methods.For more information, please visit:https://github.com/iperov/DeepFaceLab/.

cs.CV↗

arXiv2021

IB-DRR: Incremental Learning with Information-Back Discrete Representation Replay

Incremental learning aims to enable machine learning models to continuously acquire new knowledge given new classes, while maintaining the knowledge already learned for old classes. Saving a subset of training samples of previously seen classes in the memory and replaying them during new training phases is proven to be an efficient and effective way to fulfil this aim. It is evident that the larger number of exemplars the model inherits the better performance it can achieve. However, finding a trade-off between the model performance and the number of samples to save for each class is still an open problem for replay-based incremental learning and is increasingly desirable for real-life applications. In this paper, we approach this open problem by tapping into a two-step compression approach. The first step is a lossy compression, we propose to encode input images and save their discrete latent representations in the form of codes that are learned using a hierarchical Vector Quantised Variational Autoencoder (VQ-VAE). In the second step, we further compress codes losslessly by learning a hierarchical latent variable model with bits-back asymmetric numeral systems (BB-ANS). To compensate for the information lost in the first step compression, we introduce an Information Back (IB) mechanism that utilizes real exemplars for a contrastive learning loss to regularize the training of a classifier. By maintaining all seen exemplars' representations in the format of `codes', Discrete Representation Replay (DRR) outperforms the state-of-art method on CIFAR-100 by a margin of 4% accuracy with a much less memory cost required for saving samples. Incorporated with IB and saving a small set of old raw exemplars as well, the accuracy of DRR can be further improved by 2% accuracy.

cs.CV↗

arXiv2021

The critical behavior of Hegselmann-Krause opinion model with smart agents

The Hegselmann-Krause (HK) model allows one to characterize the continuous change of agents' opinions with the bounded confidence threshold $\varepsilon$. To consider the heterogeneity of agents in characteristics, we study the HK model on homogeneous and heterogeneous networks by introducing a kind of smart agent. Different from the averaging rule in opinion update of HK model, smart agents will consider, in updating their opinions, the environmental influence following the fact that an agent's behavior is often coupled with environmental changes. The environment is characterized by a parameter that represents the biased resource allocation between different cliques. We focus on the critical behavior of the underlying system. A phase transition point separating a complete consensus from the coexistence of different opinions is identified, which occurs at a critical value $\varepsilon_c$ for the bounded confidence threshold. We state analytically that $\varepsilon_c$ can take only one of two possible values, depending on the behavior of the average degree $k_a$ of a social graph, when agents are homogeneous in characteristics. Results also suggest that the phase transition point weakly depends on the network structure but is strongly correlated with the fraction of smart agents and the environmental parameter. We finally establish the finite size scaling law that stresses the role that the system size has in the underlying opinion dynamics. Meanwhile, introducing smart agents does not change the functional dependence between the time to reach a complete consensus and the system size. However, it can drive a complete consensus to be reached faster, for homogeneous networks that are far from the mean field limit.

physics.soc-ph↗

arXiv2020

Stability analysis and attractor dynamics of 3D dark solitons with localized dissipation

We study the stability and the attractor dynamics of an elongated Bose-Einstein condensate with dark or grey kink solitons in the presence of localized dissipation. To this end, the 3D Gross-Pitaevskii equation with an additional imaginary potential is solved numerically. We analyze the suppression of the snaking instability in dependence of the dissipation strength and extract the threshold value for the stabilization of the dark soliton for experimentally realistic parameters. Below the threshold value, we observe the decay into a solitonic vortex. Above the stabilization threshold, we observe the attractor dynamics towards the dark soliton when initially starting from a grey soliton. We find that for all initial conditions the dark soliton is the unique steady-state of the system - even when starting from the BEC ground state.

cond-mat.quant-gas↗

arXiv2018

Coherent perfect absorption of nonlinear matter waves

Coherent perfect absorption is the complete extinction of incoming radiation by a complex potential in a physical system supporting wave propagation. The concept was proven for linear waves in variety of systems including light interacting with absorbing scatterers, plasmonic metasurfaces and graphene films, as well as sound waves. We extend the paradigm to coherent perfect absorption of nonlinear waves and experimentally demonstrate it for matter waves in an atomic Bose-Einstein condensate. Coherent absorption of nonlinear matter waves is achieved easier than its linear analogs because the strength of two-body interactions offers additional freedom for control. Implementation of the coherent perfect absorber of Bose-Einstein condensates paves the way towards broad exploitation of the phenomenon in nonlinear optics, exciton-polariton condensates, acoustics, and other areas of nonlinear physics. It also opens perspectives for designing atomic laser.

cond-mat.quant-gas↗

arXiv2011

Hidden Tree Structure is a Key to the Emergence of Scaling in the World Wide Web

Preferential attachment is the most popular explanation for the emergence of scaling behavior in the World Wide Web, but this explanation has been challenged by the global information hypothesis, the existence of linear preference and the emergence of new big internet companies in the real world. We notice that most websites have an obvious feature that their pages are organized as a tree (namely hidden tree) and hence propose a new model that introduces a hidden tree structure into the Erdős-Ré}yi model by adding a new rule: when one node connects to another, it should also connect to all nodes in the path between these two nodes in the hidden tree. The experimental results show that the degree distribution of the generated graphs would obey power law distributions and have variable high clustering coefficients and variable small average lengths of shortest paths. The proposed model provides an alternative explanation to the emergence of scaling in the World Wide Web without the above-mentioned difficulties, and also explains the "preferential attachment" phenomenon.

cs.SI↗