Research Engineer M/F Visual Complex Anomaly Detection by Very Weakly Supervised Learning: Studies, Applications and Transposition of Visual Large Model Methods to Future Industry.

This recruitment is part of the PostGenAI project led by Sorbonne Université. The project is supported by the Agence Nationale de la Recherche (ANR), within the framework of the France 2030 government investment plan.

Your missions

• Ensure rigorous scientific study, aiming to compare the performance of VLM models on complex visual anomaly detection.

• Take charge of the complete study, dimension evaluation criteria and model deployment configurations, and conduct associated experiments.

• Contribute to writing scientific articles reporting the study's results and proposing recommendations for the transposition and deployment of VLM models in industry.

Main activities

The subject deals with the general problem of detecting complex visual anomalies in industry. Industry must equip itself with reliable inspection tools that can be quickly deployed in production. The most effective supervised AI solutions rely on a long and costly prior activity of image labeling by process experts (several hundred image labelings). Complex visual anomalies are: (1) logical defects where models must detect and control the relative positioning logic between different objects or entities in the scene. This includes, for example: checking for the absence/presence of parts or components, detecting wiring errors, etc. (2), they are also texture defects, such as opacities, cracks, tears, etc.

• The main activity of the position will be to study and then compare the performance of VLM models in detecting these anomalies. Cost criteria will be taken into account (related to inference, labeling, prompt formulations, and electricity consumption, etc.). Different types of models (open source, proprietary, etc.) will be considered for the study.

• Based on the results obtained, recommendations for the transposition of these methods to industry will be proposed and then leveraged with industrial partners of the UTC.

• Writing, submission, and publication in scientific journals are expected.

Additional information

Application Dates

From 09/21/2026 to 10/20/2026

Contract Type and Provisional Recruitment Date

Fixed-term contract - planned duration of 18 months - to be filled in October 2026

Gross Monthly Salary

According to experience and funding

Working Hours

37 hours and 30 minutes per week - 1,607 hours per year

Scientific Context / Project Summary

The PostGenAI@Paris cluster aims to anticipate breakthroughs in artificial intelligence and master their scientific, societal, and ethical challenges – where the lines between technology and human intelligence are gradually becoming indistinguishable.

https://postgenai.sorbonne-universite.fr

The project focuses on CAP 1.6 AI for Industry, which aims to develop explainability and uncertainty quantification methods to make AI usable, auditable, and truly operational in industrial decisions related to risks.

https://postgenai.sorbonne-universite.fr/pac/cap-1.6

Although there are many solutions available (local methods, global methods (Carvalho et al., 2024)), there is still room for improvement in false alarm rates. Initial findings from the benchmark study on AutoVI show that for logical defects (e.g., wiring) the false alarm rate is between 5% and 20%. It is therefore essential to introduce new methods that help explain the reasons for a false activation. Consequently, a challenge addressed in this project concerns reducing false alarm rates and improving the explainability of results for images incorrectly classified as defective.

Skills

• Ability to design and implement a rigorous experimental scientific protocol

• Solid foundation in Python programming and understanding of LLM and VLM-based AI architectures

• Knowledge of visual defect inspection for industry

Degree

Engineering degree or Master's degree

Field

Industrial Engineering, Computer Engineering

Find more English Speaking Jobs in France on Arbeitnow