arXiv subjects

Yifei Wang

Publications and source records attributed to Yifei Wang.

At least 181 records · Page 10Linked to original sources

arXiv2024

How to Craft Backdoors with Unlabeled Data Alone?

Relying only on unlabeled data, Self-supervised learning (SSL) can learn rich features in an economical and scalable way. As the drive-horse for building foundation models, SSL has received a lot of attention recently with wide applications, which also raises security concerns where backdoor attack is a major type of threat: if the released dataset is maliciously poisoned, backdoored SSL models can behave badly when triggers are injected to test samples. The goal of this work is to investigate this potential risk. We notice that existing backdoors all require a considerable amount of \emph{labeled} data that may not be available for SSL. To circumvent this limitation, we explore a more restrictive setting called no-label backdoors, where we only have access to the unlabeled data alone, where the key challenge is how to select the proper poison set without using label information. We propose two strategies for poison selection: clustering-based selection using pseudolabels, and contrastive selection derived from the mutual information principle. Experiments on CIFAR-10 and ImageNet-100 show that both no-label backdoors are effective on many SSL methods and outperform random poisoning by a large margin. Code will be available at https://github.com/PKU-ML/nlb.

cs.LG↗

arXiv2024

SM2C: Boost the Semi-supervised Segmentation for Medical Image by using Meta Pseudo Labels and Mixed Images

Recently, machine learning-based semantic segmentation algorithms have demonstrated their potential to accurately segment regions and contours in medical images, allowing the precise location of anatomical structures and abnormalities. Although medical images are difficult to acquire and annotate, semi-supervised learning methods are efficient in dealing with the scarcity of labeled data. However, overfitting is almost inevitable due to the limited images for training. Furthermore, the intricate shapes of organs and lesions in medical images introduce additional complexity in different cases, preventing networks from acquiring a strong ability to generalize. To this end, we introduce a novel method called Scaling-up Mix with Multi-Class (SM2C). This method uses three strategies - scaling-up image size, multi-class mixing, and object shape jittering - to improve the ability to learn semantic features within medical images. By diversifying the shape of the segmentation objects and enriching the semantic information within each sample, the SM2C demonstrates its potential, especially in the training of unlabelled data. Extensive experiments demonstrate the effectiveness of the SM2C on three benchmark medical image segmentation datasets. The proposed framework shows significant improvements over state-of-the-art counterparts.

cs.CV↗

arXiv2024

Do Generated Data Always Help Contrastive Learning?

Contrastive Learning (CL) has emerged as one of the most successful paradigms for unsupervised visual representation learning, yet it often depends on intensive manual data augmentations. With the rise of generative models, especially diffusion models, the ability to generate realistic images close to the real data distribution has been well recognized. These generated high-equality images have been successfully applied to enhance contrastive representation learning, a technique termed ``data inflation''. However, we find that the generated data (even from a good diffusion model like DDPM) may sometimes even harm contrastive learning. We investigate the causes behind this failure from the perspective of both data inflation and data augmentation. For the first time, we reveal the complementary roles that stronger data inflation should be accompanied by weaker augmentations, and vice versa. We also provide rigorous theoretical explanations for these phenomena via deriving its generalization bounds under data inflation. Drawing from these insights, we propose Adaptive Inflation (AdaInf), a purely data-centric strategy without introducing any extra computation cost. On benchmark datasets, AdaInf can bring significant improvements for various contrastive learning methods. Notably, without using external data, AdaInf obtains 94.70% linear accuracy on CIFAR-10 with SimCLR, setting a new record that surpasses many sophisticated methods. Code is available at https://github.com/PKU-ML/adainf.

cs.LG↗

arXiv2024

Rhythmic soliton interactions for integrated dual-microcomb spectroscopy

Rotation symmetry of microresonators supports the generation of phase-locked counter-propagating (CP) solitons that can potentially miniaturize dual-comb systems. Realization of these dual-comb compatible solitons in photonic integrated circuits remains a challenge. Here, we synthesized such CP solitons in an integrated silicon nitride microresonator and observed forced soliton oscillation due to rhythmic, time-varying soliton interactions. The interactions result in seconds mutual-coherence passively. Temporal motion in the soliton streams is discerned by measuring a quadratic-scaling frequency noise peaks and an inverse quadratic-scaling microcomb sidebands. By generating a CP soliton trimer to have two synchronized solitons in one of the orbiting directions, we resolve the incapability of measuring two unsynchronized CP soliton dimer pulses by optical cross-correlation, and show CP solitons undergo complex motion trajectory. We further prove that precise dual-comb spectroscopy with an acquisition time as short as 0.6 $μ$s is feasible using these solitons, although the temporal motion limits the dynamic range. Besides revealing soliton interactions with different group velocities, our work propels the realization of photonic integrated dual-comb spectrometers with high passive coherence.

physics.optics↗

arXiv2024

Laplacian Canonization: A Minimalist Approach to Sign and Basis Invariant Spectral Embedding

Spectral embedding is a powerful graph embedding technique that has received a lot of attention recently due to its effectiveness on Graph Transformers. However, from a theoretical perspective, the universal expressive power of spectral embedding comes at the price of losing two important invariance properties of graphs, sign and basis invariance, which also limits its effectiveness on graph data. To remedy this issue, many previous methods developed costly approaches to learn new invariants and suffer from high computation complexity. In this work, we explore a minimal approach that resolves the ambiguity issues by directly finding canonical directions for the eigenvectors, named Laplacian Canonization (LC). As a pure pre-processing method, LC is light-weighted and can be applied to any existing GNNs. We provide a thorough investigation, from theory to algorithm, on this approach, and discover an efficient algorithm named Maximal Axis Projection (MAP) that works for both sign and basis invariance and successfully canonizes more than 90% of all eigenvectors. Experiments on real-world benchmark datasets like ZINC, MOLTOX21, and MOLPCBA show that MAP consistently outperforms existing methods while bringing minimal computation overhead. Code is available at https://github.com/PKU-ML/LaplacianCanonization.

cs.LG↗

arXiv2023

Federated learning-outcome prediction with multi-layer privacy protection

Learning-outcome prediction (LOP) is a long-standing and critical problem in educational routes. Many studies have contributed to developing effective models while often suffering from data shortage and low generalization to various institutions due to the privacy-protection issue. To this end, this study proposes a distributed grade prediction model, dubbed FecMap, by exploiting the federated learning (FL) framework that preserves the private data of local clients and communicates with others through a global generalized model. FecMap considers local subspace learning (LSL), which explicitly learns the local features against the global features, and multi-layer privacy protection (MPP), which hierarchically protects the private features, including model-shareable features and not-allowably shared features, to achieve client-specific classifiers of high performance on LOP per institution. FecMap is then achieved in an iteration manner with all datasets distributed on clients by training a local neural network composed of a global part, a local part, and a classification head in clients and averaging the global parts from clients on the server. To evaluate the FecMap model, we collected three higher-educational datasets of student academic records from engineering majors. Experiment results manifest that FecMap benefits from the proposed LSL and MPP and achieves steady performance on the task of LOP, compared with the state-of-the-art models. This study makes a fresh attempt at the use of federated learning in the learning-analytical task, potentially paving the way to facilitating personalized education with privacy protection.

cs.LG↗

arXiv2023

A Machine Learning Approach to Robustly Determine Director Fields and Analyze Defects in Active Nematics

Active nematics are dense systems of rodlike particles that consume energy to drive motion at the level of the individual particles. They exist in natural systems like biological tissues and artificial materials such as suspensions of self-propelled colloidal particles or synthetic microswimmers. Active nematics have attracted significant attention in recent years due to their spectacular nonequilibrium collective spatiotemporal dynamics, which may enable applications in fields such as robotics, drug delivery, and materials science. The director field, which measures the direction and degree of alignment of the local nematic orientation, is a crucial characteristic of active nematic and is essential for studying topological defects. However, determining the director field is a significant challenge in many experimental systems. Although director fields can be derived from images of active nematics using traditional imaging processing methods, the accuracy of such methods are highly sensitive to the settings of the algorithms. These settings must be tuned from image-to-image due to experimental noise, intrinsic noise of the imaging technology, and perturbations caused by changes in experimental conditions. This sensitivity currently limits automatic analysis of active nematics. To address this, we developed a machine learning model for extracting reliable director fields from raw experimental images, which enables accurate analysis of topological defects. Application of the algorithm to experimental data demonstrates that the approach is robust and highly generalizable to experimental settings that are different from those in the training data. It could be a promising tool for investigating active nematics and may be generalized to other active matter systems.

cond-mat.soft↗

arXiv2023

Polynomial-Time Solutions for ReLU Network Training: A Complexity Classification via Max-Cut and Zonotopes

We investigate the complexity of training a two-layer ReLU neural network with weight decay regularization. Previous research has shown that the optimal solution of this problem can be found by solving a standard cone-constrained convex program. Using this convex formulation, we prove that the hardness of approximation of ReLU networks not only mirrors the complexity of the Max-Cut problem but also, in certain special cases, exactly corresponds to it. In particular, when $ε\leq\sqrt{84/83}-1\approx 0.006$, we show that it is NP-hard to find an approximate global optimizer of the ReLU network objective with relative error $ε$ with respect to the objective value. Moreover, we develop a randomized algorithm which mirrors the Goemans-Williamson rounding of semidefinite Max-Cut relaxations. To provide polynomial-time approximations, we classify training datasets into three categories: (i) For orthogonal separable datasets, a precise solution can be obtained in polynomial-time. (ii) When there is a negative correlation between samples of different classes, we give a polynomial-time approximation with relative error $\sqrt{π/2}-1\approx 0.253$. (iii) For general datasets, the degree to which the problem can be approximated in polynomial-time is governed by a geometric factor that controls the diameter of two zonotopes intrinsic to the dataset. To our knowledge, these results present the first polynomial-time approximation guarantees along with first hardness of approximation results for regularized ReLU networks.

cs.LG↗

arXiv2023

Architecture Matters: Uncovering Implicit Mechanisms in Graph Contrastive Learning

With the prosperity of contrastive learning for visual representation learning (VCL), it is also adapted to the graph domain and yields promising performance. However, through a systematic study of various graph contrastive learning (GCL) methods, we observe that some common phenomena among existing GCL methods that are quite different from the original VCL methods, including 1) positive samples are not a must for GCL; 2) negative samples are not necessary for graph classification, neither for node classification when adopting specific normalization modules; 3) data augmentations have much less influence on GCL, as simple domain-agnostic augmentations (e.g., Gaussian noise) can also attain fairly good performance. By uncovering how the implicit inductive bias of GNNs works in contrastive learning, we theoretically provide insights into the above intriguing properties of GCL. Rather than directly porting existing VCL methods to GCL, we advocate for more attention toward the unique architecture of graph learning and consider its implicit influence when designing GCL methods. Code is available at https: //github.com/PKU-ML/ArchitectureMattersGCL.

cs.LG↗

arXiv2023

Balance, Imbalance, and Rebalance: Understanding Robust Overfitting from a Minimax Game Perspective

Adversarial Training (AT) has become arguably the state-of-the-art algorithm for extracting robust features. However, researchers recently notice that AT suffers from severe robust overfitting problems, particularly after learning rate (LR) decay. In this paper, we explain this phenomenon by viewing adversarial training as a dynamic minimax game between the model trainer and the attacker. Specifically, we analyze how LR decay breaks the balance between the minimax game by empowering the trainer with a stronger memorization ability, and show such imbalance induces robust overfitting as a result of memorizing non-robust features. We validate this understanding with extensive experiments, and provide a holistic view of robust overfitting from the dynamics of both the two game players. This understanding further inspires us to alleviate robust overfitting by rebalancing the two players by either regularizing the trainer's capacity or improving the attack strength. Experiments show that the proposed ReBalanced Adversarial Training (ReBAT) can attain good robustness and does not suffer from robust overfitting even after very long training. Code is available at https://github.com/PKU-ML/ReBAT.

cs.LG↗

arXiv2023

Identifiable Contrastive Learning with Automatic Feature Importance Discovery

Existing contrastive learning methods rely on pairwise sample contrast $z_x^\top z_{x'}$ to learn data representations, but the learned features often lack clear interpretability from a human perspective. Theoretically, it lacks feature identifiability and different initialization may lead to totally different features. In this paper, we study a new method named tri-factor contrastive learning (triCL) that involves a 3-factor contrast in the form of $z_x^\top S z_{x'}$, where $S=\text{diag}(s_1,\dots,s_k)$ is a learnable diagonal matrix that automatically captures the importance of each feature. We show that by this simple extension, triCL can not only obtain identifiable features that eliminate randomness but also obtain more interpretable features that are ordered according to the importance matrix $S$. We show that features with high importance have nice interpretability by capturing common classwise features, and obtain superior performance when evaluated for image retrieval using a few features. The proposed triCL objective is general and can be applied to different contrastive learning methods like SimCLR and CLIP. We believe that it is a better alternative to existing 2-factor contrastive learning by improving its identifiability and interpretability with minimal overhead. Code is available at https://github.com/PKU-ML/Tri-factor-Contrastive-Learning.

cs.CV↗

arXiv2023

Towards Control-Centric Representations in Reinforcement Learning from Images

Image-based Reinforcement Learning is a practical yet challenging task. A major hurdle lies in extracting control-centric representations while disregarding irrelevant information. While approaches that follow the bisimulation principle exhibit the potential in learning state representations to address this issue, they still grapple with the limited expressive capacity of latent dynamics and the inadaptability to sparse reward environments. To address these limitations, we introduce ReBis, which aims to capture control-centric information by integrating reward-free control information alongside reward-specific knowledge. ReBis utilizes a transformer architecture to implicitly model the dynamics and incorporates block-wise masking to eliminate spatiotemporal redundancy. Moreover, ReBis combines bisimulation-based loss with asymmetric reconstruction loss to prevent feature collapse in environments with sparse rewards. Empirical studies on two large benchmarks, including Atari games and DeepMind Control Suit, demonstrate that ReBis has superior performance compared to existing methods, proving its effectiveness.

cs.LG↗

arXiv2023

Digital Twin System for Home Service Robot Based on Motion Simulation

In order to improve the task execution capability of home service robot, and to cope with the problem that purely physical robot platforms cannot sense the environment and make decisions online, a method for building digital twin system for home service robot based on motion simulation is proposed. A reliable mapping of the home service robot and its working environment from physical space to digital space is achieved in three dimensions: geometric, physical and functional. In this system, a digital space-oriented URDF file parser is designed and implemented for the automatic construction of the robot geometric model. Next, the physical model is constructed from the kinematic equations of the robot and an improved particle swarm optimization algorithm is proposed for the inverse kinematic solution. In addition, to adapt to the home environment, functional attributes are used to describe household objects, thus improving the semantic description of the digital space for the real home environment. Finally, through geometric model consistency verification, physical model validity verification and virtual-reality consistency verification, it shows that the digital twin system designed in this paper can construct the robot geometric model accurately and completely, complete the operation of household objects successfully, and the digital twin system is effective and practical.

cs.RO↗

arXiv2023

Robust Long-Tailed Learning via Label-Aware Bounded CVaR

Data in the real-world classification problems are always imbalanced or long-tailed, wherein the majority classes have the most of the samples that dominate the model training. In such setting, the naive model tends to have poor performance on the minority classes. Previously, a variety of loss modifications have been proposed to address the long-tailed leaning problem, while these methods either treat the samples in the same class indiscriminatingly or lack a theoretical guarantee. In this paper, we propose two novel approaches based on CVaR (Conditional Value at Risk) to improve the performance of long-tailed learning with a solid theoretical ground. Specifically, we firstly introduce a Label-Aware Bounded CVaR (LAB-CVaR) loss to overcome the pessimistic result of the original CVaR, and further design the optimal weight bounds for LAB-CVaR theoretically. Based on LAB-CVaR, we additionally propose a LAB-CVaR with logit adjustment (LAB-CVaR-logit) loss to stabilize the optimization process, where we also offer the theoretical support. Extensive experiments on real-world datasets with long-tailed label distributions verify the superiority of our proposed methods.

cs.LG↗

arXiv2023

On fusogenicity of positively charged phased-separated lipid vesicles: experiments and computational simulations

This paper studies the fusogenicity of cationic liposomes in relation to their surface distribution of cationic lipids and utilizes membrane phase separation to control this surface distribution. It is found that concentrating the cationic lipids into small surface patches on liposomes, through phase-separation, can enhance liposome's fusogenicity. Further concentrating these lipids into smaller patches on the surface of liposomes led to an increased level of fusogenicity. These experimental findings are supported by numerical simulations using a mathematical model for phase-separated charged liposomes. Findings of this study may be used for design and development of highly fusogenic liposomes with minimal level of toxicity.

cond-mat.soft↗

arXiv2023

Thickness and temperature dependent damping in La$_{0.67}$Sr$_{0.33}$MnO$_{3}$ epitaxial films

The damping of La0.67Sr0.33MnO3 (LSMO) epitaxial films as a function of thickness at different temperatures was studied. The competition between two scattering types (\r{ho}-like and σ-like) with entirely distinct thickness and temperature dependencies resulted in complicated damping behavior. The behavior of σ-like damping in LSMO films is consistent with the behavior in magnetic metal films. However, because \r{ho}-like damping is sensitive to the fine electron structure near the Fermi surface, the distortion of the oxygen octahedra controlled by the film thickness is an important factor in controlling the damping. Our study demonstrates that the complexity of damping in LSMO epitaxial films is a consequence of strong-correlation effects, which are characteristic of complex transition-metal oxides.

cond-mat.mtrl-sci↗

arXiv2023

Rethinking Weak Supervision in Helping Contrastive Learning

Contrastive learning has shown outstanding performances in both supervised and unsupervised learning, and has recently been introduced to solve weakly supervised learning problems such as semi-supervised learning and noisy label learning. Despite the empirical evidence showing that semi-supervised labels improve the representations of contrastive learning, it remains unknown if noisy supervised information can be directly used in training instead of after manual denoising. Therefore, to explore the mechanical differences between semi-supervised and noisy-labeled information in helping contrastive learning, we establish a unified theoretical framework of contrastive learning under weak supervision. Specifically, we investigate the most intuitive paradigm of jointly training supervised and unsupervised contrastive losses. By translating the weakly supervised information into a similarity graph under the framework of spectral clustering based on the posterior probability of weak labels, we establish the downstream classification error bound. We prove that semi-supervised labels improve the downstream error bound whereas noisy labels have limited effects under such a paradigm. Our theoretical findings here provide new insights for the community to rethink the role of weak supervision in helping contrastive learning.

cs.LG↗

arXiv2023

On the Generalization of Multi-modal Contrastive Learning

Multi-modal contrastive learning (MMCL) has recently garnered considerable interest due to its superior performance in visual tasks, achieved by embedding multi-modal data, such as visual-language pairs. However, there still lack theoretical understandings of how MMCL extracts useful visual representation from multi-modal pairs, and particularly, how MMCL outperforms previous approaches like self-supervised contrastive learning (SSCL). In this paper, by drawing an intrinsic connection between MMCL and asymmetric matrix factorization, we establish the first generalization guarantees of MMCL for visual downstream tasks. Based on this framework, we further unify MMCL and SSCL by showing that MMCL implicitly performs SSCL with (pseudo) positive pairs induced by text pairs. Through this unified perspective, we characterize the advantage of MMCL by showing that text pairs induce more semantically consistent and diverse positive pairs, which, according to our analysis, provably benefit downstream generalization. Inspired by this finding, we propose CLIP-guided resampling methods to significantly improve the downstream performance of SSCL on ImageNet by leveraging multi-modal information. Code is available at https://github.com/PKU-ML/CLIP-Help-SimCLR.

cs.CV↗