EVENT

ERATO Project Colloquium by Sebastian Schneider

On Tuesday 29 October, Sebastian Schneider (Technical University of Munich) will give a talk, "Search-Based Repair of Neural Networks for Monocular Depth Estimation", for our project colloquium at 16:30. Further details can be found below.

Title:

Search-Based Repair of Neural Networks for Monocular Depth Estimation

Speaker:

Sebastian Schneider (Technical University of Munich)

Abstract:

Deep Neural Networks (DNNs) have become a core component in several safety-critical tasks like Autonomous Driving. To ensure their safe usage in these scenarios, DNN repair has been used to fix erroneous predictions by changing the values of a small subset of parameters. However, these methods require the presence of categorical outputs to work effectively. We propose a new approach called RegRep to enable practitioners to repair regression networks, which are increasingly deployed in these critical domains. We follow previous approaches by selecting suspicious weights according to their performance on a targeted repair metric. Our method then employs a search-based filtering step to ensure the quality of chosen weights before generating a patch with new value assignments for the identified weights. Experiments show a significant increase in repair capabilities compared to state-of-the-art Art repair approaches and standardized methodologies like Fine-Tuning, thus highlighting its ability to address core issues faced in industry applications.

Time/Date:

16:30- / Tuesday 29 October,2024

Place:

NII Room 1310A and Online

Link:

For the latest information about ERATO colloquium/seminar, please see the webpage.
https://docs.google.com/document/d/1Qrg4c8XDkbO3tmns6tQwxn5lGHOrBON5LtHXXTpXDeA/edit

Contact:

Kittiphon Phalakarn (ERATO MMSD Colloquium Organizer)
Email: kphalakarn[at] nii.ac.jp