Junran Yang

- Designed an outlier analysis method by iterating deep learning models and using GESD to detect false labels in marker-labeled dataset for publication on CVPR 2023 - Developed a SOTA method for multi-view multi-bird tracking in 3D space utilizing 2D keypoint estimation, multi-view geometry and multi-object tracking - Fine-tuned deep learning models for object detection, pose estimation, and multi-object tracking - Orchestrated model training and evaluation pipeline in both Kubernetes and Slurm cluster

- Experimented topic models with GuidedLDA and BERTopic, and performed named entity recognition with HuggingFace models on 760k Inetrnet articles in German - Built a text classification pipeline using LangChain with OpenAI models and instruct-tuned Llama2, benchmarked with other text classification models from HuggingFace - Organized parallel computational tasks in Slurm cluster

- Developed a package to measure user trackability under different levels of browsing history exposure - Built interactive visualizations for statistical correlation between online behaviors and demographics- Developed a package to measure user trackability under different levels of browsing history exposure - Built interactive visualizations for statistical correlation between online behaviors and demographics