This internship will explore using Video Language Models (VLMs) to improve brand safety and suitability classification, focusing on the GARM framework. Current methods like keyword analysis struggle with the complex context of video and MUM underperforms in several segments. VLMs, combining visual and textual understanding, offer a solution.

The project will:

• Understand GARM: Review GARM’s Brand Safety and Suitability Framework and its content categories.
• Explore VLMs: Investigate suitable VLM architectures and training datasets, addressing data bias.
• Evaluate Performance: Benchmark VLM performance against existing methods using GARM-defined metrics.
• Address Challenges: Consider computational cost, explainability, and potential attacks on VLM systems.

Please refer the Job description for details