Applied Researcher 1

eBay•Bellevue, WA

19h•$147,200 - $196,500•Hybrid

About The Position

eBay, Inc. seeks Applied Researcher 1 in Bellevue, WA. This role involves developing, implementing, and optimizing virtual try-on technologies using advanced generative AI models like Stable Diffusion XL (SDXL) and FLUX. The focus is on enhancing realism and image fidelity through techniques such as flow matching and transformer-based models with attention mechanisms. The researcher will collaborate with engineering and product teams for production deployment, conduct rigorous performance assessments, and contribute to research publications. Maintaining up-to-date knowledge of generative AI advancements is crucial for refining product offerings.

Requirements

Master’s degree, or foreign equivalent, in Data Science, Computer Science, Applied Mathematics, or a related quantitative discipline plus 18 months of experience as a software engineer or machine learning engineer.
Attention and self-attention mechanisms, including transformer-based designs (ViT, Swin, BERT, or LLaMA)
Architecting custom attention layers or modifying transformer backbones for vision or multimodal applications.
Scaling laws, memory optimization, and distributed attention computation.
Implementing OCR, object detection, image segmentation, and image classification pipelines.
Frameworks such as torch vision or OpenCV
Integrating visual encoders (CLIP, BLIP, Flamingo, or Kosmos) with large language models.
Large-scale data preprocessing using PySpark or similar distributed data frameworks.
Designing efficient data pipelines for model training and evaluation at scale (terabytes to petabytes).
Quantitative and qualitative evaluation methodologies for computer vision and generative models.
Machine learning validation principles, including proper dataset splitting (train/validation/test), cross-validation, and prevention of data leakage.
Designing robust evaluation pipelines, ensuring statistical significance of results, reproducibility of experiments, and proper use of control baselines or ablation studies.
Coding skills in Python, PyTorch, Transformers, and Hugging Face ecosystems.
Working with ML Ops and model deployment workflows (Docker, Ray, Airflow, or MLflow).

Responsibilities

Develop, implement, and optimize virtual try-on technologies using Stable Diffusion XL (SDXL) and FLUX, incorporating advanced sampling techniques and flow matching methodologies to enhance realism and image fidelity.
Apply flow matching techniques to improve the efficiency and stability of generative models, enabling more accurate and faster image synthesis.
Leverage transformer-based models and attention mechanisms to effectively capture complex visual patterns in virtual garment generation.
Collaborate closely with software engineering and product management teams to integrate and deploy advanced virtual try-on models into production environments.
Perform rigorous performance assessments, including quantitative analyses of model accuracy, visual realism, user experience, and stability across various diffusion settings and model configurations.
Write detailed technical documentation, comprehensive research reports, and summaries outlining methodologies, results, and guidelines to facilitate internal knowledge sharing and practical application of developed technologies.
Contribute actively to peer-reviewed research publications and technical white papers related to advanced virtual try-on methodologies, innovations in diffusion models, and transformer architectures.
Maintain updated knowledge of cutting-edge developments in generative AI, diffusion processes, and related deep learning methodologies to continuously enhance and refine virtual try-on product offerings.