The data privacy settings of your browser prevent this site from functioning properly. Please open the site in a new tab. Alternatively, you can allow cross-sitetracking in your browser.
Open in new tab
Thesis Opportunity in Medical Technology (f/m/d) - Job Friedberg, Home office - Career at Digital Endoscopy GmbH

Thesis Opportunity in Medical Technology (f/m/d)

Thesis Opportunity for Bachelor's/Master's Students


Thesis Topic: Detection of Colonic Polyps using DNN compared on a GPGPU and an Analog Matrix Processor

Endoscopic screenings are crucial for colorectal cancer prevention. During these examinations, physicians search the mucosa with an endoscope for polyps. This process demands significant attention from the doctor and can benefit from machine learning methods. Object detection, for instance, can direct the physician's attention to suspicious regions for further examination. Deep neural convolutional networks are commonly used for such tasks due to their excellent detection performance, despite being computationally intensive. Hence, AI accelerators, like the Mythic ME10761, are often utilized for inference tasks due to their efficiency.

Thesis Scope:
The Mythic ME10761 is a hybrid analog/digital processor optimized for typical neural network computations. It performs calculations analogously, quantizing results back to the digital domain. Utilizing such an accelerator requires converting a neural network’s parameters trained on a GPU to the accelerator's analog architecture and quantizing intermediate results. While Mythic provides an SDK for the former, this thesis aims to analyze the effects of quantization on detection performance. Additionally, as analog processors can theoretically yield different results for identical inputs, even with result quantization, this work will experimentally evaluate the frequency and impact of such occurrences on network performance.


Your tasks

  1. Integrate Mythic's SDK into an existing object detection training pipeline.

  2. Implement an inference pipeline integrated into an existing evaluation framework.

  3. Evaluate the accelerator's inference speed (e.g., achievable FPS).

  4. Assess the speed of parameter loading for task switching.

  5. Compare the detection performance of the converted neural network on the accelerator with the original version.

  6. Evaluate potential fluctuations in detection performance due to the accelerator's analog operation.


Your qualifications

  • Degree in Biomedical Engineering, Software Engineering with a focus on AI, Electrical Engineering, Computer Engineering, or a related field.

Exciting Opportunity in Medical Device Development


    What Awaits You:

  • Be a pioneer in medical device innovation.

  • Shape the future of endoscopy.

  • Your work can make a global impact.

  • Experience an open team culture and welcoming work environment.

  • Work in a diverse and innovative team.

  • Join a dynamic team with efficient communication channels.

  • Direct interaction with top-tier medical professionals.

  • Become part of a global organization.

  • What We Offer:

  • Permanent employment contract.

  • High level of responsibility and influence.

  • Flexible work hours with the option for remote work.

  • Enjoy complimentary drinks and coffee.

  • Subsidized public transportation pass.

  • Job bike program.

  • Company voucher program for various shops.

  • Engage in company parties and events.

  • Support provided for language courses for non-native speakers.

  • Why Join Us:

  • Engaging and stimulating tasks in an innovative environment with a wealth of knowledge.

  • Experience flat hierarchies combined with the security of a global corporation.

  • Work within a highly motivated team under optimal conditions, offering independent and challenging responsibilities.

  • Benefit from specialized training and personal development initiatives.

  • Competitive salary and excellent career progression opportunities.