A5062753670- Anomaly Detection Techniques and Applications
- Explainable Artificial Intelligence (XAI)
- Scientific Computing and Data Management
- Advanced Malware Detection Techniques
- Ethics and Social Impacts of AI
- Network Security and Intrusion Detection
- Bayesian Modeling and Causal Inference
- Data-Driven Disease Surveillance
- Software System Performance and Reliability
- Adversarial Robustness in Machine Learning
- Advanced Statistical Process Monitoring
- Decision-Making and Behavioral Economics
- Fault Detection and Control Systems
- Advanced Statistical Methods and Models
- Privacy-Preserving Technologies in Data
- Time Series Analysis and Forecasting
Worcester Polytechnic Institute
2022-2025
University at Buffalo, State University of New York
2019
Due to the scarcity of reliable anomaly labels, recent detection methods leveraging noisy auto-generated labels either select clean samples or refurbish labels. However, both approaches struggle due unique properties anomalies. Sample selection often fails separate sufficiently many from ones, while label refurbishment erroneously refurbishes marginal samples. To overcome these limitations, we design Unity, first learning (LNL) approach for that elegantly leverages merits sample and...
We are able to faithfully reproduce the original paper's findings as well key results reported in its experimental section (i.e., explanation quality, speed, robustness). The authors provided example data, comprehensive scripts, and plotting functions that allowed near-identical reconstruction of all figures.
Performing deep learning on end-user devices provides fast offline inference results and can help protect the user's privacy. However, running models untrusted client reveals model information which may be proprietary, i.e., operating system or other applications manipulated to copy redistribute this information, infringing provider's intellectual property. We propose use of ARM TrustZone, a hardware-based security feature present in most phones, confidentially run proprietary an device....
Deep Learning techniques have been widely used in detecting anomalies from complex data. Most of these are either unsupervised or semi-supervised because a lack large number labeled anomalies. However, they typically rely on clean training data not polluted by to learn the distribution normal Otherwise, learned tends be distorted and hence ineffective distinguishing between abnormal To solve this problem, we propose novel approach called ELITE that uses small examples infer hidden samples....
Machine learning systems are deployed in domains such as hiring and healthcare, where undesired classifications can have serious ramifications for the user. Thus, there is a rising demand explainable AI which provide actionable steps lay users to obtain their desired outcome. To meet this need, we propose FACET, first explanation analytics system supports user interactively refining counterfactual explanations decisions made by tree ensembles. As FACET's foundation, design novel type of...
Anomaly detection is a critical task in applications like preventing financial fraud, system malfunctions, and cybersecurity attacks. While previous research has offered plethora of anomaly algorithms, effective remains challenging for users due to the tedious manual tuning process. Currently, model developers must determine which these numerous algorithms best suited their particular domain then tune many parameters by hand make chosen algorithm perform well. This demonstration showcases...
Automated decision-making systems are increasingly deployed in domains such as hiring and credit approval where negative outcomes can have substantial ramifications for decision subjects. Thus, recent research has focused on providing explanations that help subjects understand the system enable them to take actionable recourse change their outcome. Popular counterfactual explanation techniques aim achieve this by describing alterations an instance would transform a outcome positive one....
Machine learning is routinely used to automate consequential decisions about users in domains such as finance and healthcare, raising concerns of transparency recourse for negative outcomes. Existing Explainable AI techniques generate a static counterfactual point explanation which recommends changes user's instance obtain positive outcome. Unfortunately, these recommendations are often difficult or impossible realistically enact. To overcome this, we present FACET, the first interactive...
Log anomaly detection, critical in identifying system failures and preempting security breaches, finds irregular patterns within large volumes of log data. Modern detectors rely on training deep learning models clean anomaly-free However, such data requires expensive tedious human labeling. In this paper, we thus propose a robust detection framework, PlutoNOSPACE, that automatically selects representative sample subset the polluted sequence to train Transformer-based model. Pluto features...