Microsoft is looking to advance multimodal face normalization and RGB+NIR recognition for secure sign-in experiences, such as Windows Hello, by exploring techniques that transform diverse, cross-modal inputs into robust identity representations and improve invariance to pose/illumination/occlusion.
Requirements
- Currently enrolled in a Master’s or Ph.D. program in CS, EE, Applied Math, or related field with a focus in vision/graphics/ML.
- Publications in CVPR/ICCV/ECCV/NeurIPS/ICLR/ICML/SIGGRAPH or related journals.
- Experience in face recognition/verification, multimodal learning/fusion, metric learning, representation or generative modeling.
- Depth in multimodal normalization (pre-FR normalizers, prototype learning) and RGB↔NIR FR.
- Experience with VLMs/LLMs (prompting, fine-tuning, tool-use) for visual reasoning, explainability, or safety.
- Scalable training (DDP/multi-node), dataset curation, reproducible MLOps; familiarity with liveness/FAS and fairness/robustness evaluation.
- Proficient PyTorch/JAX background; ability to implement/reproduce SOTA.
Responsibilities
- Research & prototype methods in areas such as unified face normalization across modalities (e.g., RGB↔NIR), with joint prototype + feature learning and cross-modal alignment.
- Multimodal face recognition (fusion across RGB, NIR, depth/IR, audio cues where appropriate), with robustness/fairness under distribution shift.
- Large Language Models-aided face verification: explore Vision Language Models (VLM)/Large Language Models (LLM) pipelines that use visual context in the photo to assist verification.
- Efficiency & reliability: distillation/quantization/pruning, lightweight encoders/normalizers, calibration and uncertainty, liveness/antispoof integration.
- Evaluate thoroughly: define datasets and protocols; run ablations and benchmarks (ROC, EER, TPR@FAR, latency/memory, fairness/robustness).
- Production immersion: learn Windows Hello-style pipelines (signals, constraints, on-device considerations) to align research with deployment.
- Publish: communicate results via talks, internal tech reports, and submissions to top venues.
Other
- Currently enrolled in a Master’s or Ph.D. program
- Research Interns are expected to be physically located in their manager’s Microsoft worksite location for the duration of their internship.
- Submit a minimum of two reference letters for this position as well as a cover letter and any relevant work or research samples.
- Must be able to work in the United States without requiring sponsorship
- Must be available for a minimum of 12 weeks
