A5053521358- Data Quality and Management
- Big Data and Business Intelligence
- Information and Cyber Security
- Cloud Data Security Solutions
- Open Source Software Innovations
- Advanced Image and Video Retrieval Techniques
- Anomaly Detection Techniques and Applications
- Semantic Web and Ontologies
- Video Surveillance and Tracking Methods
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Software Engineering Research
- Image Retrieval and Classification Techniques
- Time Series Analysis and Forecasting
- Mobile Crowdsensing and Crowdsourcing
- Adversarial Robustness in Machine Learning
- Bacillus and Francisella bacterial research
- Gaussian Processes and Bayesian Inference
- Video Analysis and Summarization
- Phase Equilibria and Thermodynamics
- Machine Learning in Materials Science
- High-Velocity Impact and Material Behavior
- Quantum, superfluid, helium dynamics
- Expert finding and Q&A systems
- Access Control and Trust
Amazon (United States)
2024
Rochester Institute of Technology
2019-2024
Background: As smart and automated applications pervade our lives, an increasing number of software developers are required to incorporate machine learning (ML) techniques into application development. However, acquiring the ML skill set can be nontrivial for owing both breadth depth domain. Aims: We seek understand challenges face in process development offer insights simplify process. Despite its importance, there has been little research on this topic. A few existing studies with...
Multiple-instance learning (MIL) provides an effective way to tackle the video anomaly detection problem by modeling it as a weakly supervised labels are usually only available at level while missing for frames due expensive labeling cost. We propose conduct novel Bayesian non-parametric submodular partition (BN-SVP) significantly improve MIL model training that can offer highly reliable solution robust in practical settings include outlier segments or multiple types of abnormal events....
Deep learning models with large-scale backbones have been increasingly adopted to tackle complex visual question answering (VQA) problems in real settings. While providing powerful capacities handle the high-dimensional and multimodal VQA data, these tend suffer from memorization effect leading overconfident predictions. This can significantly limit their applicability critical domains (e.g., medicine, cyber-security, public safety), where confidently wrong predictions may lead severe...
Trust between developers influences the success of open source software (OSS) projects. Although existing research recognizes importance trust, there is a lack an effective and scalable computational method to measure trust in OSS community. Consequently, project members must rely on subjective inferences based fragile incomplete information for trust-related decision making. We propose automated approach assist developer identifying trustworthiness another developer. Our two-fold approach,...
Prior neural architecture search (NAS) for adversarial robustness works have discovered that a lightweight and adversarially robust network could exist in non-robust large teacher network, generally disclosed by heuristic rules through statistical analysis search, from search. However, methods cannot uniformly handle different attacks "teacher" capacity. To solve this challenge, we propose Reinforced Compressive Neural Architecture Search (RC-NAS) Versatile Adversarial Robustness....
Prior research on neural architecture search (NAS) for adversarial robustness has revealed that a lightweight and adversarially robust sub-network could exist in non-robust large teacher network. Such is generally discovered based heuristic rules to perform search. However, are inadequate handle diverse attacks different "teacher" network capacity. To address this key challenge, we propose Reinforced Compressive Neural Architecture Search (RC-NAS), aiming achieve Versatile Adversarial...
Fine-Grained Sketch-Based Image Retrieval (FG-SBIR) aims at finding a specific image from large gallery given query sketch. Despite the widespread applicability of FG-SBIR in many critical domains (e.g., crime activity tracking), existing approaches still suffer low accuracy while being sensitive to external noises such as unnecessary strokes The retrieval performance will further deteriorate under more practical on-the-fly setting, where only partially complete sketch with few (noisy) are...
Ab initio molecular dynamics (AIMD) simulations have become an important tool used in the construction of equations state (EOS) tables for warm dense matter. Due to computational costs, only a limited number system conditions can be simulated, and remaining EOS surface must interpolated use radiation-hydrodynamic experiments. In this work, we develop thermodynamically consistent model that utilizes physics-informed machine learning approach implicitly learn underlying Helmholtz free-energy...
As a widely used weakly supervised learning scheme, modern multiple instance (MIL) models achieve competitive performance at the bag level. However, instance-level prediction, which is essential for many important applications, remains largely unsatisfactory. We propose to conduct novel active deep that samples small subset of informative instances annotation, aiming significantly boost prediction. A variance regularized loss function designed properly balance bias and predictions,...