Talk details

The world is inherently multimodal, consisting of sights, sounds, and other sensory data. To achieve a human-like understanding of this complex world, AI models require multimodal data for analysis. This talk delves into Multimodal AI, exploring why it's essential for achieving superior AI performance. We'll explore how multimodal learning approaches work, how generative AI models like Gemini can supercharge these capabilities, and best practices for maximizing the value of your multimodal AI projects. Since building and running these models can be resource-intensive, we'll discuss strategies to optimize their utilization. Additionally, we'll showcase practical examples of multimodal AI models using platforms like Google AI Studio (text-to-image, image-to-text, video-to-text) and tools like PaliGemma and Idefics2 (https://huggingface.co/docs/transformers/main/en/model_doc/idefics2).