Researcher2017 - 2021German Research Center for Artificial Intelligence (DFKI)
Technologies: Python, CUDA, TensorFlow, PyTorch
- Co-supervised bachelor's and master's projects and thesis.
- Created deep learning systems for automated fault detection in the industry (mainly anomaly detection).
- Developed table detection and structure recognition systems (mostly published as research papers).
- Demystified deep learning models for time-series analysis through visualization.
- Created adversarial examples and defenses for time-series and visual modalities.
Researcher2019 - 2019NVIDIA Research
Technologies: CUDA, PyTorch, Python
- Developed a principled approach towards defending against attacks through better understanding the root cause of the vulnerability of deep learning models against adversarial attacks.
- Employed alternate visual image representations inspired by the 3D human visual system to enhance adversarial robustness.
- Defined large-scale deep learning benchmarks with efficient data loading pipelines.
- Created attention models and weakly-supervised object localization.
Technical Project Lead2016 - 2017RheinMain University of Applied Sciences (HS-RM)
Technologies: CUDA, TensorFlow, C++, Python
- Served as the technical project lead for FibeVid project (fish biodiversity estimation by low-cost non-destructive video-based sampling).
- Developed a deep learning-based segmentation system for real-time monocular object detection and tracking.
- Led fish classification using cross-layer pooling of deep CNN.
Research Assistant2015 - 2016TUKL-NUST Research \& Development Center (SEECS, NUST)
Technologies: TensorFlow, C++, Python
- Performed document detection and analysis using OpenCV and Tesseract - ICDAR SmartDoc competition 2015 (7th Position).
- Provided document detection and classification using convolutional and LSTM networks.
- Detected and segmented objects using convolutional neural networks.
- Contributed to fish detection and classification using classical computer vision approaches coupled with deep learning.