arXiv subjects

Li Gu

Publications and source records attributed to Li Gu.

25 records · Page 2Linked to original sources

arXiv2022

Improving ProtoNet for Few-Shot Video Object Recognition: Winner of ORBIT Challenge 2022

In this work, we present the winning solution for ORBIT Few-Shot Video Object Recognition Challenge 2022. Built upon the ProtoNet baseline, the performance of our method is improved with three effective techniques. These techniques include the embedding adaptation, the uniform video clip sampler and the invalid frame detection. In addition, we re-factor and re-implement the official codebase to encourage modularity, compatibility and improved performance. Our implementation accelerates the data loading in both training and testing.

cs.CV↗

arXiv2022

Few-Shot Class-Incremental Learning via Entropy-Regularized Data-Free Replay

Few-shot class-incremental learning (FSCIL) has been proposed aiming to enable a deep learning system to incrementally learn new classes with limited data. Recently, a pioneer claims that the commonly used replay-based method in class-incremental learning (CIL) is ineffective and thus not preferred for FSCIL. This has, if truth, a significant influence on the fields of FSCIL. In this paper, we show through empirical results that adopting the data replay is surprisingly favorable. However, storing and replaying old data can lead to a privacy concern. To address this issue, we alternatively propose using data-free replay that can synthesize data by a generator without accessing real data. In observing the the effectiveness of uncertain data for knowledge distillation, we impose entropy regularization in the generator training to encourage more uncertain examples. Moreover, we propose to relabel the generated data with one-hot-like labels. This modification allows the network to learn by solely minimizing the cross-entropy loss, which mitigates the problem of balancing different objectives in the conventional knowledge distillation approach. Finally, we show extensive experimental results and analysis on CIFAR-100, miniImageNet and CUB-200 to demonstrate the effectiveness of our proposed one.

cs.CV↗

arXiv2019

DMM-Net: Differentiable Mask-Matching Network for Video Object Segmentation

In this paper, we propose the differentiable mask-matching network (DMM-Net) for solving the video object segmentation problem where the initial object masks are provided. Relying on the Mask R-CNN backbone, we extract mask proposals per frame and formulate the matching between object templates and proposals at one time step as a linear assignment problem where the cost matrix is predicted by a CNN. We propose a differentiable matching layer by unrolling a projected gradient descent algorithm in which the projection exploits the Dykstra's algorithm. We prove that under mild conditions, the matching is guaranteed to converge to the optimum. In practice, it performs similarly to the Hungarian algorithm during inference. Meanwhile, we can back-propagate through it to learn the cost matrix. After matching, a refinement head is leveraged to improve the quality of the matched mask. Our DMM-Net achieves competitive results on the largest video object segmentation dataset YouTube-VOS. On DAVIS 2017, DMM-Net achieves the best performance without online learning on the first frames. Without any fine-tuning, DMM-Net performs comparably to state-of-the-art methods on SegTrack v2 dataset. At last, our matching layer is very simple to implement; we attach the PyTorch code ($<50$ lines) in the supplementary material. Our code is released at https://github.com/ZENGXH/DMM_Net.

cs.CV↗

arXiv2018

Adversarial Distillation of Bayesian Neural Network Posteriors

Bayesian neural networks (BNNs) allow us to reason about uncertainty in a principled way. Stochastic Gradient Langevin Dynamics (SGLD) enables efficient BNN learning by drawing samples from the BNN posterior using mini-batches. However, SGLD and its extensions require storage of many copies of the model parameters, a potentially prohibitive cost, especially for large neural networks. We propose a framework, Adversarial Posterior Distillation, to distill the SGLD samples using a Generative Adversarial Network (GAN). At test-time, samples are generated by the GAN. We show that this distillation framework incurs no loss in performance on recent BNN applications including anomaly detection, active learning, and defense against adversarial attacks. By construction, our framework not only distills the Bayesian predictive distribution, but the posterior itself. This allows one to compute quantities such as the approximate model variance, which is useful in downstream tasks. To our knowledge, these are the first results applying MCMC-based BNNs to the aforementioned downstream applications.

cs.LG↗

arXiv2017

Space-Filling Designs for Robustness Experiments

To identify the robust settings of the control factors, it is very important to understand how they interact with the noise factors. In this article, we propose space-filling designs for computer experiments that are more capable of accurately estimating the control-by-noise interactions. Moreover, the existing space-filling designs focus on uniformly distributing the points in the design space, which are not suitable for noise factors because they usually follow non-uniform distributions such as normal distribution. This would suggest placing more points in the regions with high probability mass. However, noise factors also tend to have a smooth relationship with the response and therefore, placing more points towards the tails of the distribution is also useful for accurately estimating the relationship. These two opposing effects make the experimental design methodology a challenging problem. We propose optimal and computationally efficient solutions to this problem and demonstrate their advantages using simulated examples and a real industry example involving a manufacturing packing line.

stat.ME↗

arXiv2017

Deterministic Sampling of Expensive Posteriors Using Minimum Energy Designs

Markov chain Monte Carlo (MCMC) methods require a large number of samples to approximate a posterior distribution, which can be costly when the likelihood or prior is expensive to evaluate. The number of samples can be reduced if we can avoid repeated samples and those that are close to each other. This is the idea behind deterministic sampling methods such as Quasi-Monte Carlo (QMC). However, the existing QMC methods aim at sampling from a uniform hypercube, which can miss the high probability regions of the posterior distribution and thus the approximation can be poor. Minimum energy design (MED) is a recently proposed deterministic sampling method, which makes use of the posterior evaluations to obtain a weighted space-filling design in the region of interest. However, the existing implementation of MED is inefficient because it requires several global optimizations and thus numerous evaluations of the posterior. In this article, we develop an efficient algorithm that can generate MED with few posterior evaluations. We also make several improvements to the MED criterion to make it perform better in high dimensions. The advantages of MED over MCMC and QMC are illustrated using an example of calibrating a friction drilling process.

stat.CO↗

arXiv2014

Tracing The Largest Seasonal Migration on Earth

It is estimated that over 3.6 billion passengers are travelling during the Chinese Spring Festival travel season. They leave their working cities and return their hometowns to enjoy annual family time, and back to cities after the holiday. In this study, with the massive location-based data collected from millions of smartphone users, we propose a novel method to trace the migration flow and explore the migration patterns of Chinese people. From the temporal perspective, we explore the migration trend over time during a 34-days period, about half a month before and after the Spring Festival. From the spatial perspective, the migration directions and routes are estimated and quantified, and the migration flow is visualized. The spatial range of influence of developed regions could be reflected with the destinations of migration, the migration destinations and originations have obvious characteristic of geographical proximity.

physics.soc-ph↗