arXiv subjects

Zhipeng Cai

Publications and source records attributed to Zhipeng Cai.

71 records · Page 4Linked to original sources

arXiv2023

Online Continual Learning Without the Storage Constraint

Traditional online continual learning (OCL) research has primarily focused on mitigating catastrophic forgetting with fixed and limited storage allocation throughout an agent's lifetime. However, a broad range of real-world applications are primarily constrained by computational costs rather than storage limitations. In this paper, we target such applications, investigating the online continual learning problem under relaxed storage constraints and limited computational budgets. We contribute a simple algorithm, which updates a kNN classifier continually along with a fixed, pretrained feature extractor. We selected this algorithm due to its exceptional suitability for online continual learning. It can adapt to rapidly changing streams, has zero stability gap, operates within tiny computational budgets, has low storage requirements by only storing features, and has a consistency property: It never forgets previously seen data. These attributes yield significant improvements, allowing our proposed algorithm to outperform existing methods by over 20% in accuracy on two large-scale OCL datasets: Continual LOCalization (CLOC) with 39M images and 712 classes and Continual Google Landmarks V2 (CGLM) with 580K images and 10,788 classes, even when existing methods retain all previously seen images. Furthermore, we achieve this superior performance with considerably reduced computational and storage expenses. We provide code to reproduce our results at github.com/drimpossible/ACM.

cs.CV↗

arXiv2023

Graph Convolutional Network with Connectivity Uncertainty for EEG-based Emotion Recognition

Automatic emotion recognition based on multichannel Electroencephalography (EEG) holds great potential in advancing human-computer interaction. However, several significant challenges persist in existing research on algorithmic emotion recognition. These challenges include the need for a robust model to effectively learn discriminative node attributes over long paths, the exploration of ambiguous topological information in EEG channels and effective frequency bands, and the mapping between intrinsic data qualities and provided labels. To address these challenges, this study introduces the distribution-based uncertainty method to represent spatial dependencies and temporal-spectral relativeness in EEG signals based on Graph Convolutional Network (GCN) architecture that adaptively assigns weights to functional aggregate node features, enabling effective long-path capturing while mitigating over-smoothing phenomena. Moreover, the graph mixup technique is employed to enhance latent connected edges and mitigate noisy label issues. Furthermore, we integrate the uncertainty learning method with deep GCN weights in a one-way learning fashion, termed Connectivity Uncertainty GCN (CU-GCN). We evaluate our approach on two widely used datasets, namely SEED and SEEDIV, for emotion recognition tasks. The experimental results demonstrate the superiority of our methodology over previous methods, yielding positive and significant improvements. Ablation studies confirm the substantial contributions of each component to the overall performance.

cs.LG↗

arXiv2023

CLNeRF: Continual Learning Meets NeRF

Novel view synthesis aims to render unseen views given a set of calibrated images. In practical applications, the coverage, appearance or geometry of the scene may change over time, with new images continuously being captured. Efficiently incorporating such continuous change is an open challenge. Standard NeRF benchmarks only involve scene coverage expansion. To study other practical scene changes, we propose a new dataset, World Across Time (WAT), consisting of scenes that change in appearance and geometry over time. We also propose a simple yet effective method, CLNeRF, which introduces continual learning (CL) to Neural Radiance Fields (NeRFs). CLNeRF combines generative replay and the Instant Neural Graphics Primitives (NGP) architecture to effectively prevent catastrophic forgetting and efficiently update the model when new data arrives. We also add trainable appearance and geometry embeddings to NGP, allowing a single compact model to handle complex scene changes. Without the need to store historical images, CLNeRF trained sequentially over multiple scans of a changing scene performs on-par with the upper bound model trained on all scans at once. Compared to other CL baselines CLNeRF performs much better across standard benchmarks and WAT. The source code, and the WAT dataset are available at https://github.com/IntelLabs/CLNeRF. Video presentation is available at: https://youtu.be/nLRt6OoDGq0?si=8yD6k-8MMBJInQPs

cs.CV↗

arXiv2023

Metric3D: Towards Zero-shot Metric 3D Prediction from A Single Image

Reconstructing accurate 3D scenes from images is a long-standing vision task. Due to the ill-posedness of the single-image reconstruction problem, most well-established methods are built upon multi-view geometry. State-of-the-art (SOTA) monocular metric depth estimation methods can only handle a single camera model and are unable to perform mixed-data training due to the metric ambiguity. Meanwhile, SOTA monocular methods trained on large mixed datasets achieve zero-shot generalization by learning affine-invariant depths, which cannot recover real-world metrics. In this work, we show that the key to a zero-shot single-view metric depth model lies in the combination of large-scale data training and resolving the metric ambiguity from various camera models. We propose a canonical camera space transformation module, which explicitly addresses the ambiguity problems and can be effortlessly plugged into existing monocular models. Equipped with our module, monocular models can be stably trained with over 8 million images with thousands of camera models, resulting in zero-shot generalization to in-the-wild images with unseen camera settings. Experiments demonstrate SOTA performance of our method on 7 zero-shot benchmarks. Notably, our method won the championship in the 2nd Monocular Depth Estimation Challenge. Our method enables the accurate recovery of metric 3D structures on randomly collected internet images, paving the way for plausible single-image metrology. The potential benefits extend to downstream tasks, which can be significantly improved by simply plugging in our model. For example, our model relieves the scale drift issues of monocular-SLAM (Fig. 1), leading to high-quality metric scale dense mapping. The code is available at https://github.com/YvanYin/Metric3D.

cs.CV↗

arXiv2023

Random-Order Enumeration for Self-Reducible NP-Problems

In plenty of data analysis tasks, a basic and time-consuming process is to produce a large number of solutions and feed them into downstream processing. Various enumeration algorithms have been developed for this purpose. An enumeration algorithm produces all solutions of a problem instance without repetition. To be a statistically meaningful representation of the solution space, solutions are required to be enumerated in uniformly random order. This paper studies a set of self-reducible NP-problems in three hierarchies, where the problems are polynomially countable ($Sr_{NP}^{FP}$), admit FPTAS ($Sr_{NP}^{FPTAS}$), and admit FPRAS ($Sr_{NP}^{FPRAS}$), respectively. The trivial algorithm based on a (almost) uniform generator is in fact inefficient. We provide a new insight that the (almost) uniform generator is not the end of the story. More efficient algorithmic frameworks are proposed to enumerate solutions in uniformly random order for problems in these three hierarchies. (1) For problems in $Sr_{NP}^{FP}$, we show a random-order enumeration algorithm with polynomial delay (PDREnum); (2) For problems in $Sr_{NP}^{FPTAS}$, we show a Las Vegas random-order enumeration algorithm with expected polynomial delay (PDLVREnum); (3) For problems in $Sr_{NP}^{FPRAS}$, we devise a fully polynomial delay Atlantic City random-order enumeration algorithm with expected delay polynomial in the input size and the given error probability $δ$ (FPACREnum), which has a probability of at least $1-δ$ becoming a Las Vegas random-order enumeration algorithm. Finally, to further improve the efficiency of the random-order enumeration algorithms, based on the master/slave paradigm, we present a parallelization with 1.5-optimal enumeration delay and running time, along with the theoretical analysis.

cs.DS↗

arXiv2022

Improving information retention in large scale online continual learning

Given a stream of data sampled from non-stationary distributions, online continual learning (OCL) aims to adapt efficiently to new data while retaining existing knowledge. The typical approach to address information retention (the ability to retain previous knowledge) is keeping a replay buffer of a fixed size and computing gradients using a mixture of new data and the replay buffer. Surprisingly, the recent work (Cai et al., 2021) suggests that information retention remains a problem in large scale OCL even when the replay buffer is unlimited, i.e., the gradients are computed using all past data. This paper focuses on this peculiarity to understand and address information retention. To pinpoint the source of this problem, we theoretically show that, given limited computation budgets at each time step, even without strict storage limit, naively applying SGD with constant or constantly decreasing learning rates fails to optimize information retention in the long term. We propose using a moving average family of methods to improve optimization for non-stationary objectives. Specifically, we design an adaptive moving average (AMA) optimizer and a moving-average-based learning rate schedule (MALR). We demonstrate the effectiveness of AMA+MALR on large-scale benchmarks, including Continual Localization (CLOC), Google Landmarks, and ImageNet. Code will be released upon publication.

cs.CV↗

arXiv2021

Online Continual Learning with Natural Distribution Shifts: An Empirical Study with Visual Data

Continual learning is the problem of learning and retaining knowledge through time over multiple tasks and environments. Research has primarily focused on the incremental classification setting, where new tasks/classes are added at discrete time intervals. Such an "offline" setting does not evaluate the ability of agents to learn effectively and efficiently, since an agent can perform multiple learning epochs without any time limitation when a task is added. We argue that "online" continual learning, where data is a single continuous stream without task boundaries, enables evaluating both information retention and online learning efficacy. In online continual learning, each incoming small batch of data is first used for testing and then added to the training set, making the problem truly online. Trained models are later evaluated on historical data to assess information retention. We introduce a new benchmark for online continual visual learning that exhibits large scale and natural distribution shifts. Through a large-scale analysis, we identify critical and previously unobserved phenomena of gradient-based optimization in continual learning, and propose effective strategies for improving gradient-based online continual learning with real data. The source code and dataset are available in: https://github.com/IntelLabs/continuallearning.

cs.LG↗

arXiv2021

Generative Adversarial Networks: A Survey Towards Private and Secure Applications

Generative Adversarial Networks (GAN) have promoted a variety of applications in computer vision, natural language processing, etc. due to its generative model's compelling ability to generate realistic examples plausibly drawn from an existing distribution of samples. GAN not only provides impressive performance on data generation-based tasks but also stimulates fertilization for privacy and security oriented research because of its game theoretic optimization strategy. Unfortunately, there are no comprehensive surveys on GAN in privacy and security, which motivates this survey paper to summarize those state-of-the-art works systematically. The existing works are classified into proper categories based on privacy and security functions, and this survey paper conducts a comprehensive analysis of their advantages and drawbacks. Considering that GAN in privacy and security is still at a very initial stage and has imposed unique challenges that are yet to be well addressed, this paper also sheds light on some potential privacy and security applications with GAN and elaborates on some future research directions.

cs.LG↗

arXiv2020

Collaborative City Digital Twin For Covid-19 Pandemic: A Federated Learning Solution

In this work, we propose a collaborative city digital twin based on FL, a novel paradigm that allowing multiple city DT to share the local strategy and status in a timely manner. In particular, an FL central server manages the local updates of multiple collaborators (city DT), provides a global model which is trained in multiple iterations at different city DT systems, until the model gains the correlations between various response plan and infection trend. That means, a collaborative city DT paradigm based on FL techniques can obtain knowledge and patterns from multiple DTs, and eventually establish a `global view' for city crisis management. Meanwhile, it also helps to improve each city digital twin selves by consolidating other DT's respective data without violating privacy rules. To validate the proposed solution, we take COVID-19 pandemic as a case study. The experimental results on the real dataset with various response plan validate our proposed solution and demonstrate the superior performance.

cs.LG↗

arXiv2020

Adversarial Privacy Preserving Graph Embedding against Inference Attack

Recently, the surge in popularity of Internet of Things (IoT), mobile devices, social media, etc. has opened up a large source for graph data. Graph embedding has been proved extremely useful to learn low-dimensional feature representations from graph structured data. These feature representations can be used for a variety of prediction tasks from node classification to link prediction. However, existing graph embedding methods do not consider users' privacy to prevent inference attacks. That is, adversaries can infer users' sensitive information by analyzing node representations learned from graph embedding algorithms. In this paper, we propose Adversarial Privacy Graph Embedding (APGE), a graph adversarial training framework that integrates the disentangling and purging mechanisms to remove users' private information from learned node representations. The proposed method preserves the structural information and utility attributes of a graph while concealing users' private attributes from inference attacks. Extensive experiments on real-world graph datasets demonstrate the superior performance of APGE compared to the state-of-the-arts. Our source code can be found at https://github.com/uJ62JHD/Privacy-Preserving-Social-Network-Embedding.

cs.LG↗

arXiv2020

Complexity and Efficient Algorithms for Data Inconsistency Evaluating and Repairing

Data inconsistency evaluating and repairing are major concerns in data quality management. As the basic computing task, optimal subset repair is not only applied for cost estimation during the progress of database repairing, but also directly used to derive the evaluation of database inconsistency. Computing an optimal subset repair is to find a minimum tuple set from an inconsistent database whose remove results in a consistent subset left. Tight bound on the complexity and efficient algorithms are still unknown. In this paper, we improve the existing complexity and algorithmic results, together with a fast estimation on the size of optimal subset repair. We first strengthen the dichotomy for optimal subset repair computation problem, we show that it is not only APXcomplete, but also NPhard to approximate an optimal subset repair with a factor better than $17/16$ for most cases. We second show a $(2-0.5^{\tinyσ-1})$-approximation whenever given $σ$ functional dependencies, and a $(2-η_k+\frac{η_k}{k})$-approximation when an $η_k$-portion of tuples have the $k$-quasi-Tur$\acute{\text{a}}$n property for some $k>1$. We finally show a sublinear estimator on the size of optimal \textit{S}-repair for subset queries, it outputs an estimation of a ratio $2n+εn$ with a high probability, thus deriving an estimation of FD-inconsistency degree of a ratio $2+ε$. To support a variety of subset queries for FD-inconsistency evaluation, we unify them as the $\subseteq$-oracle which can answer membership-query, and return $p$ tuples uniformly sampled whenever given a number $p$. Experiments are conducted on range queries as an implementation of $\subseteq$-oracle, and results show the efficiency of our FD-inconsistency degree estimator.

cs.DB↗

arXiv2019

Consensus Maximization Tree Search Revisited

Consensus maximization is widely used for robust fitting in computer vision. However, solving it exactly, i.e., finding the globally optimal solution, is intractable. A* tree search, which has been shown to be fixed-parameter tractable, is one of the most efficient exact methods, though it is still limited to small inputs. We make two key contributions towards improving A* tree search. First, we show that the consensus maximization tree structure used previously actually contains paths that connect nodes at both adjacent and non-adjacent levels. Crucially, paths connecting non-adjacent levels are redundant for tree search, but they were not avoided previously. We propose a new acceleration strategy that avoids such redundant paths. In the second contribution, we show that the existing branch pruning technique also deteriorates quickly with the problem dimension. We then propose a new branch pruning technique that is less dimension-sensitive to address this issue. Experiments show that both new techniques can significantly accelerate A* tree search, making it reasonably efficient on inputs that were previously out of reach.

cs.CV↗

arXiv2018

Deterministic consensus maximization with biconvex programming

Consensus maximization is one of the most widely used robust fitting paradigms in computer vision, and the development of algorithms for consensus maximization is an active research topic. In this paper, we propose an efficient deterministic optimization algorithm for consensus maximization. Given an initial solution, our method conducts a deterministic search that forcibly increases the consensus of the initial solution. We show how each iteration of the update can be formulated as an instance of biconvex programming, which we solve efficiently using a novel biconvex optimization algorithm. In contrast to our algorithm, previous consensus improvement techniques rely on random sampling or relaxations of the objective function, which reduce their ability to significantly improve the initial consensus. In fact, on challenging instances, the previous techniques may even return a worse off solution. Comprehensive experiments show that our algorithm can consistently and greatly improve the quality of the initial solution, without substantial cost.

cs.CV↗

arXiv2018

Practical optimal registration of terrestrial LiDAR scan pairs

Point cloud registration is a fundamental problem in 3D scanning. In this paper, we address the frequent special case of registering terrestrial LiDAR scans (or, more generally, levelled point clouds). Many current solutions still rely on the Iterative Closest Point (ICP) method or other heuristic procedures, which require good initializations to succeed and/or provide no guarantees of success. On the other hand, exact or optimal registration algorithms can compute the best possible solution without requiring initializations; however, they are currently too slow to be practical in realistic applications. Existing optimal approaches ignore the fact that in routine use the relative rotations between scans are constrained to the azimuth, via the built-in level compensation in LiDAR scanners. We propose a novel, optimal and computationally efficient registration method for this 4DOF scenario. Our approach operates on candidate 3D keypoint correspondences, and contains two main steps: (1) a deterministic selection scheme that significantly reduces the candidate correspondence set in a way that is guaranteed to preserve the optimal solution; and (2) a fast branch-and-bound (BnB) algorithm with a novel polynomial-time subroutine for 1D rotation search, that quickly finds the optimal alignment for the reduced set. We demonstrate the practicality of our method on realistic point clouds from multiple LiDAR surveys.

cs.CV↗

arXiv2018

Robust Fitting in Computer Vision: Easy or Hard?

Robust model fitting plays a vital role in computer vision, and research into algorithms for robust fitting continues to be active. Arguably the most popular paradigm for robust fitting in computer vision is consensus maximisation, which strives to find the model parameters that maximise the number of inliers. Despite the significant developments in algorithms for consensus maximisation, there has been a lack of fundamental analysis of the problem in the computer vision literature. In particular, whether consensus maximisation is "tractable" remains a question that has not been rigorously dealt with, thus making it difficult to assess and compare the performance of proposed algorithms, relative to what is theoretically achievable. To shed light on these issues, we present several computational hardness results for consensus maximisation. Our results underline the fundamental intractability of the problem, and resolve several ambiguities existing in the literature.

cs.CV↗

arXiv2016

Technique Report: Scheduling Flows with Multiple Service Frequency Constraints

With the fast development of wireless technologies, wireless applications have invaded various areas in people's lives with a wide range of capabilities. Guaranteeing Quality-of-Service (QoS) is the key to the success of those applications. One of the QoS requirements, service frequency, is very important for tasks including multimedia transmission in the Internet of Things. A service frequency constraint denotes the length of time period during which a link can transmit at least once. Unfortunately, it has not been well addressed yet. Therefore, this paper proposes a new framework to schedule multi transmitting flows in wireless networks considering service frequency constraint for each link. In our model, the constraints for flows are heterogeneous due to the diversity of users' behaviors. We first introduce a new definition for network stability with service frequency constraints and demonstrate that the novel scheduling policy is throughput-optimal in one fundamental category of network models. After that, we discuss the performance of a wireless network with service frequency constraints from the views of capacity region and total queue length. Finally, a series of evaluations indicate the proposed scheduling policy can guarantee service frequency and achieve a good performance on the aspect of queue length of each flow.

cs.NI↗

arXiv2015

Exploring the genetic patterns of complex diseases via the integrative genome-wide approach

Motivation: Genome-wide association studies (GWASs), which assay more than a million single nucleotide polymorphisms (SNPs) in thousands of individuals, have been widely used to identify genetic risk variants for complex diseases. However, most of the variants that have been identified contribute relatively small increments of risk and only explain a small portion of the genetic variation in complex diseases. This is the so-called missing heritability problem. Evidence has indicated that many complex diseases are genetically related, meaning these diseases share common genetic risk variants. Therefore, exploring the genetic correlations across multiple related studies could be a promising strategy for removing spurious associations and identifying underlying genetic risk variants, and thereby uncovering the mystery of missing heritability in complex diseases. Results: We present a general and robust method to identify genetic patterns from multiple large-scale genomic datasets. We treat the summary statistics as a matrix and demonstrate that genetic patterns will form a low-rank matrix plus a sparse component. Hence, we formulate the problem as a matrix recovering problem, where we aim to discover risk variants shared by multiple diseases/traits and those for each individual disease/trait. We propose a convex formulation for matrix recovery and an efficient algorithm to solve the problem. We demonstrate the advantages of our method using both synthesized datasets and real datasets. The experimental results show that our method can successfully reconstruct both the shared and the individual genetic patterns from summary statistics and achieve better performance compared with alternative methods under a wide range of scenarios.

cs.CE↗