arXiv subjects

Yajing Liu

Publications and source records attributed to Yajing Liu.

At least 19 recordsLinked to original sources

arXiv2026

A UCB Bandit Algorithm for General ML-Based Estimators

We present ML-UCB, a generalized upper confidence bound algorithm that integrates arbitrary machine learning models into multi-armed bandit frameworks. A fundamental challenge in deploying sophisticated ML models for sequential decision-making is the lack of tractable concentration inequalities required for principled exploration. We overcome this limitation by directly modeling the learning curve behavior of the underlying estimator. Specifically, assuming the Mean Squared Error decreases as a power law in the number of training samples, we derive a generalized concentration inequality and prove that ML-UCB achieves sublinear regret. This framework enables the principled integration of any ML model whose learning curve can be empirically characterized, eliminating the need for model-specific theoretical analysis. We validate our approach through experiments on a collaborative filtering recommendation system using online matrix factorization with synthetic data designed to simulate a simplified two-tower model, demonstrating substantial improvements over LinUCB

cs.LG↗

arXiv2025

LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions

Scheduling virtual machines (VMs) on hosts in cloud data centers dictates efficiency and is an NP-hard problem with incomplete information. Prior work improved VM scheduling with predicted VM lifetimes. Our work further improves lifetime-aware scheduling using repredictions with lifetime distributions versus one-shot prediction. Our approach repredicts and adjusts VM and host lifetimes when incorrect predictions emerge. We also present novel approaches for defragmentation and regular system maintenance, which are essential to our data center reliability and optimizations, and are not explored in prior work. We show repredictions deliver a fundamental advance in effectiveness over one-shot prediction. We call our novel combination of distribution-based lifetime predictions and scheduling algorithms Lifetime Aware VM Allocation (LAVA). LAVA reduces resource stranding and increases the number of empty hosts, which are critical for large VM scheduling, cloud system updates, and reducing dynamic energy consumption. Our approach runs in production within Google's hyperscale cloud data centers, where it improves efficiency by decreasing stranded compute and memory resources by ~3% and ~2% respectively. It increases empty hosts by 2.3-9.2 pp in production, reducing dynamic energy consumption, and increasing availability for large VMs and cloud system updates. We also show a reduction in VM migrations for host defragmentation and maintenance. In addition to our fleet-wide production deployment, we perform simulation studies to characterize the design space and show that our algorithm significantly outperforms the prior state of the art lifetime-based scheduling approach.

cs.DC↗

arXiv2024

Unbiased Faster R-CNN for Single-source Domain Generalized Object Detection

Single-source domain generalization (SDG) for object detection is a challenging yet essential task as the distribution bias of the unseen domain degrades the algorithm performance significantly. However, existing methods attempt to extract domain-invariant features, neglecting that the biased data leads the network to learn biased features that are non-causal and poorly generalizable. To this end, we propose an Unbiased Faster R-CNN (UFR) for generalizable feature learning. Specifically, we formulate SDG in object detection from a causal perspective and construct a Structural Causal Model (SCM) to analyze the data bias and feature bias in the task, which are caused by scene confounders and object attribute confounders. Based on the SCM, we design a Global-Local Transformation module for data augmentation, which effectively simulates domain diversity and mitigates the data bias. Additionally, we introduce a Causal Attention Learning module that incorporates a designed attention invariance loss to learn image-level features that are robust to scene confounders. Moreover, we develop a Causal Prototype Learning module with an explicit instance constraint and an implicit prototype constraint, which further alleviates the negative impact of object attribute confounders. Experimental results on five scenes demonstrate the prominent generalization ability of our method, with an improvement of 3.9% mAP on the Night-Clear scene.

cs.CV↗

arXiv2023

Efficient current-induced spin torques and field-free magnetization switching in a room-temperature van der Waals magnet

The discovery of magnetism in van der Waals (vdW) materials has established unique building blocks for the research of emergent spintronic phenomena. In particular, owing to their intrinsically clean surface without dangling bonds, the vdW magnets hold the potential to construct a superior interface that allows for efficient electrical manipulation of magnetism. Despite several attempts in this direction, it usually requires a cryogenic condition and the assistance of external magnetic fields, which is detrimental to the real application. Here, we fabricate heterostructures based on Fe3GaTe2 flakes that possess room-temperature ferromagnetism with excellent perpendicular magnetic anisotropy. The current-driven non-reciprocal modulation of coercive fields reveals a high spin-torque efficiency in the Fe3GaTe2/Pt heterostructures, which further leads to a full magnetization switching by current. Moreover, we demonstrate the field-free magnetization switching resulting from out-of-plane polarized spin currents by asymmetric geometry design. Our work could expedite the development of efficient vdW spintronic logic, memory and neuromorphic computing devices.

cond-mat.mtrl-sci↗

arXiv2023

ReLU Neural Networks, Polyhedral Decompositions, and Persistent Homolog

A ReLU neural network leads to a finite polyhedral decomposition of input space and a corresponding finite dual graph. We show that while this dual graph is a coarse quantization of input space, it is sufficiently robust that it can be combined with persistent homology to detect homological signals of manifolds in the input space from samples. This property holds for a variety of networks trained for a wide range of purposes that have nothing to do with this topological application. We found this feature to be surprising and interesting; we hope it will also be useful.

math.AT↗

arXiv2023

Hamming Similarity and Graph Laplacians for Class Partitioning and Adversarial Image Detection

Researchers typically investigate neural network representations by examining activation outputs for one or more layers of a network. Here, we investigate the potential for ReLU activation patterns (encoded as bit vectors) to aid in understanding and interpreting the behavior of neural networks. We utilize Representational Dissimilarity Matrices (RDMs) to investigate the coherence of data within the embedding spaces of a deep neural network. From each layer of a network, we extract and utilize bit vectors to construct similarity scores between images. From these similarity scores, we build a similarity matrix for a collection of images drawn from 2 classes. We then apply Fiedler partitioning to the associated Laplacian matrix to separate the classes. Our results indicate, through bit vector representations, that the network continues to refine class detectability with the last ReLU layer achieving better than 95\% separation accuracy. Additionally, we demonstrate that bit vectors aid in adversarial image detection, again achieving over 95\% accuracy in separating adversarial and non-adversarial images using a simple classifier.

cs.CV↗

arXiv2022

Dual Graphs of Polyhedral Decompositions for the Detection of Adversarial Attacks

Previous work has shown that a neural network with the rectified linear unit (ReLU) activation function leads to a convex polyhedral decomposition of the input space. These decompositions can be represented by a dual graph with vertices corresponding to polyhedra and edges corresponding to polyhedra sharing a facet, which is a subgraph of a Hamming graph. This paper illustrates how one can utilize the dual graph to detect and analyze adversarial attacks in the context of digital images. When an image passes through a network containing ReLU nodes, the firing or non-firing at a node can be encoded as a bit ($1$ for ReLU activation, $0$ for ReLU non-activation). The sequence of all bit activations identifies the image with a bit vector, which identifies it with a polyhedron in the decomposition and, in turn, identifies it with a vertex in the dual graph. We identify ReLU bits that are discriminators between non-adversarial and adversarial images and examine how well collections of these discriminators can ensemble vote to build an adversarial image detector. Specifically, we examine the similarities and differences of ReLU bit vectors for adversarial images, and their non-adversarial counterparts, using a pre-trained ResNet-50 architecture. While this paper focuses on adversarial digital images, ResNet-50 architecture, and the ReLU activation function, our methods extend to other network architectures, activation functions, and types of datasets.

cs.CV↗

arXiv2022

Self-Supervision Can Be a Good Few-Shot Learner

Existing few-shot learning (FSL) methods rely on training with a large labeled dataset, which prevents them from leveraging abundant unlabeled data. From an information-theoretic perspective, we propose an effective unsupervised FSL method, learning representations with self-supervision. Following the InfoMax principle, our method learns comprehensive representations by capturing the intrinsic structure of the data. Specifically, we maximize the mutual information (MI) of instances and their representations with a low-bias MI estimator to perform self-supervised pre-training. Rather than supervised pre-training focusing on the discriminable features of the seen classes, our self-supervised model has less bias toward the seen classes, resulting in better generalization for unseen classes. We explain that supervised pre-training and self-supervised pre-training are actually maximizing different MI objectives. Extensive experiments are further conducted to analyze their FSL performance with various training settings. Surprisingly, the results show that self-supervised pre-training can outperform supervised pre-training under the appropriate conditions. Compared with state-of-the-art FSL methods, our approach achieves comparable performance on widely used FSL benchmarks without any labels of the base classes.

cs.CV↗

arXiv2022

Source-Free Domain Adaptation for Real-world Image Dehazing

Deep learning-based source dehazing methods trained on synthetic datasets have achieved remarkable performance but suffer from dramatic performance degradation on real hazy images due to domain shift. Although certain Domain Adaptation (DA) dehazing methods have been presented, they inevitably require access to the source dataset to reduce the gap between the source synthetic and target real domains. To address these issues, we present a novel Source-Free Unsupervised Domain Adaptation (SFUDA) image dehazing paradigm, in which only a well-trained source model and an unlabeled target real hazy dataset are available. Specifically, we devise the Domain Representation Normalization (DRN) module to make the representation of real hazy domain features match that of the synthetic domain to bridge the gaps. With our plug-and-play DRN module, unlabeled real hazy images can adapt existing well-trained source networks. Besides, the unsupervised losses are applied to guide the learning of the DRN module, which consists of frequency losses and physical prior losses. Frequency losses provide structure and style constraints, while the prior loss explores the inherent statistic property of haze-free images. Equipped with our DRN module and unsupervised loss, existing source dehazing models are able to dehaze unlabeled real hazy images. Extensive experiments on multiple baselines demonstrate the validity and superiority of our method visually and quantitatively.

cs.CV↗

arXiv2022

Multi-Area Distribution System State Estimation via Distributed Tensor Completion

This paper proposes a model-free distribution system state estimation method based on tensor completion using canonical polyadic decomposition. In particular, we consider a setting where the network is divided into multiple areas. The measured physical quantities at buses located in the same area are processed by an area controller. A three-way tensor is constructed to collect these measured quantities. The measurements are analyzed locally to recover the full state information of the network. A distributed closed-form iterative algorithm based on the alternating direction method of multipliers is developed to obtain the low-rank factors of the whole network state tensor where information exchange happens only between neighboring areas. The convergence properties of the distributed algorithm and the sufficient conditions on the number of samples for each smaller network that guarantee the identifiability of the factors of the state tensor are presented. To demonstrate the efficacy of the proposed algorithm and to check the identifiability conditions, numerical simulations are carried out using the IEEE 123-bus system.

math.OC↗

arXiv2022

Prompt Distribution Learning

We present prompt distribution learning for effectively adapting a pre-trained vision-language model to address downstream recognition tasks. Our method not only learns low-bias prompts from a few samples but also captures the distribution of diverse prompts to handle the varying visual representations. In this way, we provide high-quality task-related content for facilitating recognition. This prompt distribution learning is realized by an efficient approach that learns the output embeddings of prompts instead of the input embeddings. Thus, we can employ a Gaussian distribution to model them effectively and derive a surrogate loss for efficient training. Extensive experiments on 12 datasets demonstrate that our method consistently and significantly outperforms existing methods. For example, with 1 sample per category, it relatively improves the average result by 9.1% compared to human-crafted prompts.

cs.CV↗

arXiv2022

Improving Fraud Detection via Hierarchical Attention-based Graph Neural Network

Graph neural networks (GNN) have emerged as a powerful tool for fraud detection tasks, where fraudulent nodes are identified by aggregating neighbor information via different relations. To get around such detection, crafty fraudsters resort to camouflage via connecting to legitimate users (i.e., relation camouflage) or providing seemingly legitimate feedbacks (i.e., feature camouflage). A wide-spread solution reinforces the GNN aggregation process with neighbor selectors according to original node features. This method may carry limitations when identifying fraudsters not only with the relation camouflage, but with the feature camouflage making them hard to distinguish from their legitimate neighbors. In this paper, we propose a Hierarchical Attention-based Graph Neural Network (HA-GNN) for fraud detection, which incorporates weighted adjacency matrices across different relations against camouflage. This is motivated in the Relational Density Theory and is exploited for forming a hierarchical attention-based graph neural network. Specifically, we design a relation attention module to reflect the tie strength between two nodes, while a neighborhood attention module to capture the long-range structural affinity associated with the graph. We generate node embeddings by aggregating information from local/long-range structures and original node features. Experiments on three real-world datasets demonstrate the effectiveness of our model over the state-of-the-arts.

cs.LG↗

arXiv2021

Terahertz birefringent gratings for filtering and dispersion compensation

The recent development of the terahertz waveguide makes it an excellent platform for integrating many intriguing functionalities, which offers tremendous potential to build compact and robust terahertz systems. In the context of next-generation high-speed communication links at the terahertz band, engineering the dispersion and birefringence of terahertz waves is essential. Here, we experimentally demonstrate subwavelength birefringent waveguide gratings based on the low-loss cyclic olefin copolymer exploiting micro-machining fabrication techniques. Asymmetric cross-section and periodic-structural modulation along propagation direction are introduced to achieve birefringent THz grating for filtering and dispersion compensation. Because of strong index modulation in the subwavelength fiber, a high negative group velocity dispersion of -188 (-88) ps/mm/THz is achieved at 0.15 THz for x-polarization (y-polarization), i.e., 7.5 times increase compared to the state-of-the-art reported to date. Such high negative dispersion is realized in a 43 mm grating length, which is less than half of the length reported until now (e.g., 100 mm). Further, the subwavelength fiber grating filters two orthogonal polarization states and exhibits transmission dips with 8.5-dB and 7.5-dB extinction ratios for x and y polarization, respectively. Our experiment demonstrates the feasibility of using polymer-based terahertz gratings as a dispersion compensator in terahertz communications and steering polarized terahertz radiations for polarization-sensitive THz systems.

physics.optics↗

arXiv2021

Phoneme-based Distribution Regularization for Speech Enhancement

Existing speech enhancement methods mainly separate speech from noises at the signal level or in the time-frequency domain. They seldom pay attention to the semantic information of a corrupted signal. In this paper, we aim to bridge this gap by extracting phoneme identities to help speech enhancement. Specifically, we propose a phoneme-based distribution regularization (PbDr) for speech enhancement, which incorporates frame-wise phoneme information into speech enhancement network in a conditional manner. As different phonemes always lead to different feature distributions in frequency, we propose to learn a parameter pair, i.e. scale and bias, through a phoneme classification vector to modulate the speech enhancement network. The modulation parameter pair includes not only frame-wise but also frequency-wise conditions, which effectively map features to phoneme-related distributions. In this way, we explicitly regularize speech enhancement features by recognition vectors. Experiments on public datasets demonstrate that the proposed PbDr module can not only boost the perceptual quality for speech enhancement but also the recognition accuracy of an ASR system on the enhanced speech. This PbDr module could be readily incorporated into other speech enhancement networks as well.

eess.AS↗

arXiv2020

A Single Sensor Based Multispectral Imaging Camera using a Narrow Spectral Band Colour Mosaic Integrated on the Monochrome CMOS Image Sensor

A multispectral image camera captures image data within specific wavelength ranges in narrow wavelength bands across the electromagnetic spectrum. Images from a multispectral camera can extract additional information that the human eye or a normal camera fails to capture and thus may have important applications in precision agriculture, forestry, medicine and object identification. Conventional multispectral cameras are made up of multiple image sensors each fitted with a narrow passband wavelength filter and optics, which makes them heavy, bulky, power hungry and very expensive. The multiple optics also create image co-registration problem. Here, we demonstrate a single sensor based three band multispectral camera using a narrow spectral band RGB colour mosaic in a Bayer pattern integrated on a monochrome CMOS sensor. The narrow band colour mosaic is made of a hybrid combination of plasmonic colour filters and heterostructured dielectric multilayer. The demonstrated camera technology has reduced cost, weight, size and power by almost n times (where n is the number of bands) compared to a conventional multispectral camera.

physics.optics↗

arXiv2020

Hybrid colour filters for multispectral imaging

Multispectral cameras capture images in multiple wavelengths in narrow spectral bands. They offer advanced sensing well beyond normal cameras and many single sensor based multispectral cameras have been commercialized aimed at a broad range of applications, such as agroforestry research, medical analysis and so on. However, the existing single sensor based multispectral cameras require accurate alignment to overlay each filter on image sensor pixels, which makes their fabrication very complex, especially when the number of bands is large. This paper demonstrates a new filter technology using a hybrid combination of single plasmonic layer and dielectric layers by computational simulations. A filter mosaic of various bands with narrow spectral width can be achieved with single run manufacturing processes (i.e., exposure, development, deposition and other minor steps), regardless of the number of bands.

physics.optics↗

arXiv2020

A General Framework for Bounding Approximate Dynamic Programming Schemes

For years, there has been interest in approximation methods for solving dynamic programming problems, because of the inherent complexity in computing optimal solutions characterized by Bellman's principle of optimality. A wide range of approximate dynamic programming (ADP) methods now exists. It is of great interest to guarantee that the performance of an ADP scheme be at least some known fraction, say $β$, of optimal. This paper introduces a general approach to bounding the performance of ADP methods, in this sense, in the stochastic setting. The approach is based on new results for bounding greedy solutions in string optimization problems, where one has to choose a string (ordered set) of actions to maximize an objective function. This bounding technique is inspired by submodularity theory, but submodularity is not required for establishing bounds. Instead, the bounding is based on quantifying certain notions of curvature of string functions; the smaller the curvatures the better the bound. The key insight is that any ADP scheme is a greedy scheme for some surrogate string objective function that coincides in its optimal solution and value with those of the original optimal control problem. The ADP scheme then yields to the bounding technique mentioned above, and the curvatures of the surrogate objective determine the value $β$ of the bound. The surrogate objective and its curvatures depend on the specific ADP.

math.OC↗

arXiv2020

Matrix Completion Using Alternating Minimization for Distribution System State Estimation

This paper examines the problem of state estimation in power distribution systems under low-observability conditions. The recently proposed constrained matrix completion method which combines the standard matrix completion method and power flow constraints has been shown to be effective in estimating voltage phasors under low-observability conditions using single-snapshot information. However, the method requires solving a semidefinite programming (SDP) problem, which becomes computationally infeasible for large systems and if multiple-snapshot (time-series) information is used. This paper proposes an efficient algorithm to solve the constrained matrix completion problem with time-series data. This algorithm is based on reformulating the matrix completion problem as a bilinear (non-convex) optimization problem, and applying the alternating minimization algorithm to solve this problem. This paper proves the summable convergence of the proposed algorithm, and demonstrates its efficacy and scalability via IEEE 123-bus system and a real utility feeder system. This paper also explores the value of adding more data from the history in terms of computation time and estimation accuracy.

math.OC↗