Master Thesis Foundation Model Knowledge Distillation into Deployable Perce at Bosch Group

Renningen, Baden-Württemberg, Germany -

Full Time

Start Date

Immediate

Expiry Date

06 Jan, 26

Salary

0.0

Posted On

08 Oct, 25

Experience

0 year(s) or above

Remote Job

Yes

Telecommute

Yes

Sponsor Visa

Skills

Deep Learning, Computer Vision, 3D Perception Systems, Python, PyTorch, Open3D, PCL, NumPy, Knowledge Distillation, Feature Extraction, Occupancy Prediction, Object Detection, Model Evaluation, Benchmarking, Intrinsic Motivation, Project Management

Industry

Software Development

Description

Company Description At Bosch, we shape the future by inventing high-quality technologies and services that spark enthusiasm and enrich people’s lives. Our promise to our associates is rock-solid: we grow together, we enjoy our work, and we inspire each other. Join in and feel the difference. The Robert Bosch GmbH is looking forward to your application! Job Description Deep neural networks represent the state-of-the-art for 3D scene understanding in autonomous driving, including occupancy prediction or object detection. However, deployment constraints require compact models that can operate efficiently on target hardware under real-time conditions. While large foundation models have demonstrated exceptional capabilities across domains, their computational demands make direct deployment in autonomous driving challenging. This master thesis explores leveraging foundation models to enhance efficient, deployable perception architectures for 3D occupancy prediction. The objective is to effectively transfer rich knowledge from large-scale foundation models into lightweight networks suitable for real-world autonomous driving deployment. During your thesis you will investigate and compare multiple knowledge transfer strategies. Using the Knowledge Distillation approach you will implement teacher-student frameworks where foundation models guide the training of compact deployable networks. With the Foundation Model Pretraining technique you will research and develop methodologies for leveraging foundation models as sophisticated feature extractors for pretraining target architecture components. Furthermore, you will systematically implement, evaluate, and benchmark these approaches across accuracy, computational efficiency, and inference speed. The goal is to achieve substantial performance improvements while successfully bridging the gap between powerful foundation models and practical autonomous driving applications. Qualifications Education: Master studies in the field of Machine Learning, Computer Science, Math, Statistics, Physics, Cybernetics, Electrical / Mechanical Engineering or comparable with very good grades Experience and Knowledge: strong knowledge of and practical experience in deep learning, computer vision, and 3D perception systems; experience in Python as well as very good knowledge of a deep learning framework (preferably PyTorch) and data processing libraries (e.g., Open3D, PCL, NumPy) Personality and Working Practice: you independently drive tasks with strong intrinsic motivation, efficiently structuring and overseeing projects Languages: fluent in English Additional Information Start: according to prior agreement Duration: 6 months Requirement for this thesis is the enrollment at university. Please attach your CV, transcript of records, examination regulations and if indicated a valid work and residence permit. Diversity and inclusion are not just trends for us but are firmly anchored in our corporate culture. Therefore, we welcome all applications, regardless of gender, age, disability, religion, ethnic origin or sexual identity. Need further information about the job? Marc Sons (Functional Department) +49 152 2847 9248 #LI-DNI

Responsibilities

The master thesis involves exploring knowledge distillation techniques to enhance deployable perception architectures for 3D occupancy prediction in autonomous driving. The candidate will implement and benchmark various knowledge transfer strategies to improve model performance and efficiency.