A5001103949- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Multimodal Machine Learning Applications
- Adversarial Robustness in Machine Learning
- 3D Shape Modeling and Analysis
- Anomaly Detection Techniques and Applications
- Human Pose and Action Recognition
- 3D Surveying and Cultural Heritage
- Topic Modeling
- Generative Adversarial Networks and Image Synthesis
- Advanced Vision and Imaging
- Machine Learning and Data Classification
- Cancer-related molecular mechanisms research
- Machine Learning and Algorithms
- COVID-19 diagnosis using AI
- Privacy-Preserving Technologies in Data
- Genomics and Chromatin Dynamics
- Advanced Image and Video Retrieval Techniques
- RNA and protein synthesis mechanisms
- Ethics and Social Impacts of AI
- Natural Language Processing Techniques
- RNA modifications and cancer
- Bacillus and Francisella bacterial research
- Time Series Analysis and Forecasting
- Retinal Imaging and Analysis
Singapore University of Technology and Design
2025
Agency for Science, Technology and Research
2018-2024
Institute for Infocomm Research
2009-2024
National University of Singapore
2024
Stanford University
2007-2022
Generative adversarial networks (GANs) are able to model the complex highdimensional distributions of real-world data, which suggests they could be effective for anomaly detection. However, few works have explored use GANs detection task. We leverage recently developed GAN models detection, and achieve state-of-the-art performance on image network intrusion datasets, while being several hundred-fold faster at test time than only published GAN-based method.
Anomaly detection is a significant and hence well-studied problem. However, developing effective anomaly methods for complex high-dimensional data remains challenge. As Generative Adversarial Networks (GANs) are able to model the distributions of real-world data, they offer promising approach address this In work, we propose an method, Adversarially Learned Detection (ALAD) based on bi-directional GANs, that derives adversarially learned features task. ALAD then uses reconstruction errors...
Several techniques for multivariate time series anomaly detection have been proposed recently, but a systematic comparison on common set of datasets and metrics is lacking. This paper presents comprehensive evaluation unsupervised semi-supervised deep-learning based methods diagnosis data from cyberphysical systems. Unlike previous works, we vary the model post-processing errors, i.e. scoring functions independently each other, through grid 10 models 4 functions, comparing these variants to...
Enabling precise forecasting of the remaining useful life (RUL) for machines can reduce maintenance cost, increase availability, and prevent catastrophic consequences. Data-driven RUL prediction methods have already achieved acclaimed performance. However, they usually assume that training testing data are collected from same condition (same distribution or domain), which is generally not valid in real industry. Conventional approaches to address domain shift problems attempt derive...
Owing to their connection with generative adversarial networks (GANs), saddle-point problems have recently attracted considerable interest in machine learning and beyond. By necessity, most theoretical guarantees revolve around convex-concave (or even linear) problems; however, making inroads towards efficient GAN training depends crucially on moving beyond this classic framework. To make piecemeal progress along these lines, we analyze the behavior of mirror descent (MD) a class...
Unsupervised domain adaptation methods aim to generalize well on unlabeled test data that may have a different (shifted) distribution from the training data. Such are typically developed image data, and their application time series is less explored. Existing works suffer inconsistencies in evaluation schemes, datasets, backbone neural network architectures. Moreover, labeled target often used for model selection, which violates fundamental assumption of unsupervised adaptation. To address...
We examine two different techniques for parameter averaging in GAN training. Moving Average (MA) computes the time-average of parameters, whereas Exponential (EMA) an exponentially discounted sum. Whilst MA is known to lead convergence bilinear settings, we provide -- our knowledge first theoretical arguments support EMA. show that EMA converges limit cycles around equilibrium with vanishing amplitude as discount approaches one simple games and also enhances stability general establish...
State-of-the-art methods for semantic segmentation are based on deep neural networks that known to be data-hungry. Region-based active learning has shown a promising method reducing data annotation costs. A key design choice region-based AL is whether use regularly-shaped regions (e.g., rectangles) or irregularly-shaped region superpixels). In this work, we address question under realistic, click-based measurement of particular, revisit the super-pixels and demonstrate inappropriate cost...
Multiprotein complexes play central roles in many cellular pathways. Although high-throughput experimental techniques have already enabled systematic screening of pairwise protein-protein interactions en masse, the amount experimentally determined protein complex data has remained relatively lacking. As such, researchers begun to exploit vast interaction help discover new complexes. However, mining for networks is not an easy task because there are artefacts underlying due limitations...
Abstract Motivation: The need for accurate and efficient tools computational RNA structure analysis has become increasingly apparent over the last several years: folding algorithms underlie numerous applications in bioinformatics, ranging from microarray probe selection to de novo non-coding gene prediction. In this work, we present RAF (RNA Alignment Folding), an algorithm simultaneous alignment consensus of unaligned sequences. Algorithmically, exploits sparsity set likely pairing...
We consider the problem of zero-shot anomaly detection in which a model is pre-trained to detect anomalies images belonging seen classes, and expected from unseen classes at test time. State-of-the-art (AD) methods can often achieve exceptional results when training are abundant, but they catastrophically fail scenarios with lack real examples. However, emergence multi-modal models such as CLIP, it possible use knowledge other modalities (e.g. text) compensate for visual information improve...
Knowledge distillation is a learning paradigm for boosting resource-efficient graph neural networks (GNNs) using more expressive yet cumbersome teacher models. Past work on GNNs proposed the Local Structure Preserving loss (LSP), which matches local structural relationships defined over edges across student and teacher's node embeddings. This paper studies whether preserving global topology of how embeds data can be effective objective GNNs, as real-world graphs often contain latent...
Existing approaches towards anomaly detection (AD) often rely on a substantial amount of anomaly-free data to train representation and density models. However, large datasets may not always be available before the inference stage; in which case an model must trained with only handful normal samples, a.k.a. few-shot (FSAD). In this paper, we propose novel methodology address challenge FSAD incorporates two important techniques. Firstly, employ pre-trained source dataset initialize weights....
Many learning algorithms rely on the curvature (in particular, strong convexity) of regularized objective functions to provide good theoretical performance guarantees. In practice, choice regularization penalty that gives best testing set may result in with little or even no curvature. these cases, designed specifically for objectives often either fail completely require some modification involves a substantial compromise performance.
Modern convolutional object detectors have improved the detection accuracy significantly, which in turn inspired development of dedicated hardware accelerators to achieve real-time performance by exploiting inherent parallelism algorithm. Non-maximum suppression (NMS) is an indispensable operation detection. In stark contrast most operations, commonly-adopted GreedyNMS algorithm does not foster parallelism, can be a major bottleneck. this paper, we introduce MaxpoolNMS, parallelizable...
This work empirically investigates punctuation insertions as adversarial attacks on NLP systems. Data from experiments three tasks, five datasets, and six models with four show that insertions, when limited to a few symbols (apostrophes hyphens), are superior attack vector compared character due 1) lower after-attack accuracy (Aaft-atk) than alphabetical insertions; 2) higher semantic similarity between the resulting original texts; 3) text is easier faster read assessed Test of Word Reading...
A fundamental issue in deep learning has been adversarial robustness. As these systems have scaled, such issues persisted. Currently, large language models (LLMs) with billions of parameters suffer from attacks just like their earlier, smaller counterparts. However, the threat changed. Previously, having gray-box access, where input embeddings or output logits/probabilities were visible to user, might reasonable. introduction closed-source models, no information about model is available...