Presentation information coming soon!

Presentation information coming soon!

Sound is one of the world’s richest but least accessible data sources. Most audio exists as a single, tangled mix—impossible to edit for humans and noisy for machines to understand. This talk explores how advances in audio separation are unmixing the world: splitting sound into its component parts so it can be edited, searched, and programmed like code. For humans, this means new creative control, immersive experiences, and copyright-safe workflows. For machines, it provides the fine-grained data needed to power multimodal training, boost speech recognition, and enable entirely new AI applications. By transforming raw sound into structured, usable data, separation is becoming the intelligence layer for audio—laying the foundation for the next generation of media, AI, and human-computer interaction.

Generative AI is reshaping product experiences while introducing new privacy challenges. This talk explores how Meta’s Privacy-Aware Infrastructure (PAI) helps navigate this evolving landscape, protecting user data while enabling safe GenAI product innovation at scale, using Meta AI glasses as an example. We’ll highlight core PAI technologies such as data lineage and safeguards that embed privacy directly into the product development lifecycle, empowering product teams to move fast while delivering trusted experiences to Meta’s ~4bn users.

Drena Kusari (Microsoft), Mayank Ahuja (LinkedIn), Jonathan Bergeron (Meta), Abhishek Gulati (Meta), Moderated by Joseph Laria (Meta)

In this presentation, we will talk about Nvidia’s super res network called RTX VSR. Primarily created to enhance the visual experience of video playback, this technology is being widely used today in our GeForce products. We will talk about the challenges in productizing the network. We will talk about merits of this technology, show subjective visual quality improvements. Also, we will talk about how this technology is enabling a diverse set of applications where bandwidth is severely limited.

In this presentation, we will talk abut how Meta's Media Foundation team deployed advanced video super resolution (VSR) technology at scale to enhance video quality across Meta's platforms and multiple use cases.

We will explain Meta's comprehensive super resolution strategy, which applies AI-powered enhancement both server-side during video ingest and client-side during playback to improve resolution and visual quality.

We will than discuss two use cases where we deployed VSR for ads videos/images and Generative AI video workflows.
We will conclude with key learnings from leveraging large-scale subjective evaluation and objective metrics (UVQ) to measure quality improvements and detect artifacts, guiding model selection and deployment.

This session will discuss the importance of real-time communication for natural, responsive, and realistic interactions with voice AI agents, with a focus on high-bandwidth, low-latency delivery and extensibility for custom AI-driven workflows.

Attendees will learn how to:
Connect to any LLM with low-latency
Implement real-time architecture for the transport layer
Create more natural interaction with interruption handling and turn detection

This session will empower you to start building more engaging voice AI agents for use cases ranging from customer service to teaching assistants.

Advancements in conversational AI are transforming real-time communication (RTC). We've enhanced our RTC stack with new audio functionalities to facilitate natural, human-to-bot conversations. A critical aspect of this development is the effective suppression of irrelevant side speech, noise, and echo, which is vital for an always-on AI companion to avoid interruptions from background sounds or concurrent conversations. While humans instinctively differentiate foreground and background speech, generative AI bots necessitate deliberate system design to achieve this. Our approach tackles this challenge comprehensively, prioritizing distraction minimization and seamless bot interaction.

Invisible watermarking (IW) is a powerful media processing technique that enables embedding of a small amount of data into media that's imperceptible to humans but detectable by software. This technology offers a robust solution for content provenance tagging, enabling the identification of the source of content to support various use cases. This talk covers scaling an IW solution from a research model to a production deployment, adapting to challenges and constraints presented by real-world use cases.

Video conferencing technology is on the cusp of significant change using AI. We give examples how ML-based super-resolution can improve the video quality or reduce the bandwidth of traditional DSP-based video codecs by 50%. ML-based video codecs can more radically reduce the required bandwidth by 10X. Photorealistic avatars can even more radically reduce the required bandwidth by 100X and fix issues like eye gaze, video quality, and even improve how you look. Finally we discuss how generative video can be used to implement photorealistic avatars and meeting environments.

We cover the journey of Feed Deep Dive - AI-enabled experiences on Facebook Feed - from launch to finding product market fit. We look into the role of context engineering, reasoning models and agentic AI in driving user delight. We also look into cohort analysis, targeting ML models in building retentive behaviors and setting the path toward deeper usage for AI products.

Whitney Curry (Meta), Jeff Brown (Meta), Baiyang Liu (Meta)