Applied Scientist - Generative Modeling and Computer Vision

About the Role

We seek a seasoned Applied Scientist specialized in generative modeling and computer vision to join a leading Applied AI team. This position involves designing advanced diffusion-based machine learning models, leading the translation of latest AI research to production-grade applications, and mentoring a team of engineers. The role bridges pioneering research with tangible impact in products.

Key Responsibilities

• Architect, train, and fine-tune large-scale diffusion models (e.g., DDPM, DDIM, LDM, DiT) tailored to image, video, and multimodal content generation.
• Enhance sampling efficiency through methods such as distillation, consistency modeling, progressive training, and guided generation.
• Continuously assimilate and prototype emerging AI research innovations.
• Develop production-ready pipelines for computer vision tasks like segmentation, detection, depth estimation, optical flow, and 3D reconstructions.
• Fine-tune vision foundation models including Vision Transformers (ViT), CLIP, DINOv2, and SAM, utilizing parameter-efficient tuning techniques like LoRA and adapters.
• Integrate vision encoders with generative architectures to enable controllable generation capabilities (e.g., ControlNet, IP-Adapter, inpainting, editing).
• Manage the complete machine learning lifecycle encompassing data preparation, experiment tracking, model assessment, optimization, deployment, and continuous monitoring.
• Optimize model inference using techniques like INT8/FP8 quantization, ONNX exports, Flash Attention, and xFormers.
• Design and maintain scalable distributed training infrastructure leveraging technologies such as DDP, FSDP, and DeepSpeed across large GPU clusters.
• Establish evaluation frameworks and benchmarks, including human preference studies (RLHF / DPO), to appraise generative model quality.
• Lead technical design reviews, author engineering RFCs, and uphold quality standards within the team.
• Mentor junior and mid-level machine learning engineers through code reviews, one-on-one meetings, and pair programming.
• Collaborate cross-functionally with product, research, and infrastructure teams to operationalize research breakthroughs into deployed features.

Required Qualifications

• Over 15 years of direct machine learning engineering experience in industrial or research environments.
• MS or PhD degree in Computer Science, Machine Learning, Statistics, or equivalent practical expertise.
• Expert proficiency in Python programming, with robust experience in PyTorch as mandatory.
• Comprehensive theoretical and applied understanding of score-based and diffusion generative models.
• Strong foundational knowledge of computer vision concepts including Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), feature pyramids, and multi-scale processing.
• Proven experience in fine-tuning extensive vision and generative models at large scale.
• Competent with distributed training frameworks such as Distributed Data Parallel (DDP), Fully Sharded Data Parallel (FSDP), DeepSpeed, and Megatron-LM.
• Solid background in probabilistic machine learning, variational inference, and information theory.
• Experience utilizing MLOps tools like Weights & Biases, MLflow, DVC, or similar platforms.
• Demonstrated success in deploying machine learning models to production environments at scale.
• Excellent verbal and written communication skills for effective collaboration with diverse stakeholders.

Preferred Qualifications

• Experience with flow-based generative models such as normalizing flows, continuous normalizing flows (CNFs), Rectified Flow, and Flow Matching.
• Expertise in video generation architectures (e.g., Sora, CogVideo, AnimateDiff, SVD).
• Familiarity with three-dimensional generative models including Neural Radiance Fields (NeRF), 3D Gaussian Splatting, Zero-1-to-3, and Point-E.
• Background in multimodal systems combining large language models with vision (e.g., GPT-4V, LLaVA, InstructBLIP).
• Knowledge of reinforcement learning with human feedback (RLHF) and Direct Preference Optimization (DPO) approaches for aligning generative models.
• Active participation in open-source contributions, including maintained repositories or substantial pull requests to projects like HuggingFace Diffusers, CompVis, and timm.
• Ongoing engagement with the machine learning community demonstrated through an active public GitHub profile.

Equal Opportunity and Company Culture

This employer is committed to diversity and inclusion, providing equal opportunity regardless of race, religion, color, gender, sexual orientation, age, veteran or disability status, and other protected categories. They foster a workplace culture designed to empower employees to innovate and impact the company's growth trajectory.

The organization boasts a global workforce dedicated to empowering creativity and delivering innovative AI-powered experiences via renowned software products. They encourage candidates passionate about integrating AI in impactful ways to join their mission-driven team.

Interview Guidelines and Accessibility

Interviews focus on evaluating genuine skills and thought processes. Use of AI tools or recording during live interviews is prohibited unless explicitly allowed for accommodations. Applicants requiring assistance navigating the application or interview process due to disabilities can request support via provided contact points.