A5045190437- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Semantic Web and Ontologies
- Time Series Analysis and Forecasting
- Adversarial Robustness in Machine Learning
- Topic Modeling
- Multimodal Machine Learning Applications
- Cell Image Analysis Techniques
- Natural Language Processing Techniques
- Soft Robotics and Applications
- Machine Learning and ELM
- Optical Imaging and Spectroscopy Techniques
- Imbalanced Data Classification Techniques
- Face Recognition and Perception
- Functional Brain Connectivity Studies
- Tactile and Sensory Interactions
- Visual Attention and Saliency Detection
- Transportation and Mobility Innovations
- Autonomous Vehicle Technology and Safety
- AI-based Problem Solving and Planning
- Advanced Sensor and Energy Harvesting Materials
- Advanced Computational Techniques and Applications
Nvidia (United States)
2025
Carnegie Mellon University
2018-2021
Abstract Vision science, particularly machine vision, has been revolutionized by introducing large-scale image datasets and statistical learning approaches. Yet, human neuroimaging studies of visual perception still rely on small numbers images (around 100) due to time-constrained experimental procedures. To apply approaches that include neuroscience, the number used in must be significantly increased. We present BOLD5000, a functional MRI (fMRI) study includes almost 5,000 distinct...
Soft tactile skins can provide an in-depth understanding of contact location and force through a soft deformable interface. However, widespread implementation robotic sensing remains limited due to non-scalable fabrication techniques, lack customization, complex integration requirements. In this work, we demonstrate magnetic composites fabricated with two different matrix materials, silicone elastomer urethane foam, that be used as continuous surfaces for single-point localization. Building...
Recently, promising progress has been made by open-source vision-language models (VLMs) in bringing their capabilities closer to those of proprietary frontier models. However, most only publish final model weights, leaving the critical details data strategies and implementation largely opaque. In this work, we address VLM post-training from a data-centric perspective, showing key role strategy developing VLMs. By studying building our scratch, share detailed insights into development...
The advances in multimodal large language models (MLLMs) have led to growing interests LLM-based autonomous driving agents leverage their strong reasoning capabilities. However, capitalizing on MLLMs' capabilities for improved planning behavior is challenging since requires full 3D situational awareness beyond 2D reasoning. To address this challenge, our work proposes a holistic framework alignment between agent and tasks. Our starts with novel MLLM architecture that uses sparse queries lift...
In recent years, the data collected for artificial intelligence has grown to an unmanageable amount. Particularly within industrial applications, such as autonomous vehicles, model training computation budgets are being exceeded while performance is saturating -- and yet more continues pour in. To navigate flood of data, we propose a framework select most semantically diverse important dataset portion. Then, further enrich it by discovering meaningful new from massive unlabeled pool....
Methods in long-tail learning focus on improving performance for data-poor (rare) classes; however, such classes remains much lower than more data-rich (frequent) classes. Analyzing the predictions of methods rare reveals that a large number errors are due to misclassification items as visually similar frequent To address this problem, we introduce AlphaNet, method can be applied existing models, performing post hoc correction classifiers Starting with pre-trained model, find closest model's...
Large-scale datasets are essential to modern day deep learning. Advocates argue that understanding these methods requires dataset transparency (e.g. "dataset curation, motivation, composition, collection process, etc..."). However, almost no one has suggested the release of detailed definitions and visual category examples provided annotators - information critical structure annotations present in each dataset. These labels at heart public datasets, yet few include instructions were used...
Training on datasets with long-tailed distributions has been challenging for major recognition tasks such as classification and detection. To deal this challenge, image resampling is typically introduced a simple but effective approach. However, we observe that detection differs from since multiple classes may be present in one image. As result, alone not enough to yield sufficiently balanced distribution at the object level. We address object-level by introducing an object-centric memory...