NVIDIA GTC May 2020 Keynote Pt5: NVIDIA Jarvis for Conversational AI | Summary and Q&A

TL;DR

NVIDIA introduces Jarvis, a multimodal conversational AI service framework, enabling interactive and low-latency conversational AI applications.

Key Insights

• 🎯 TensorRT, NVIDIA's optimizing compiler, addresses the challenge of compiling complex computational graphs into target machines efficiently.
• 😑 Jarvis simplifies the development of conversational AI services by providing pre-trained models and a tool, Nemo, for fine-tuning with domain-specific data.
• ✊ The introduction of interactive 3D chatbots powered by Jarvis enhances the conversational AI experience, enabling realistic and engaging interactions.
• 🎮 Jarvis is a game-changer for real-time inference applications like video conferencing, offering features such as simultaneous speech recognition, closed captions, translations, and summarizations.
• ❤️‍🩹 NVIDIA's AI breakthroughs, including TensorRT, Merlin (for recommender systems), and Jarvis, form an end-to-end acceleration platform for AI applications.
• 💗 The increased adoption of TensorRT and the success of Jarvis highlight the growing importance and impact of deep learning and AI in various industries.
• 🐕‍🦺 Use of NVIDIA GPUs in data centers for inference by the world's top 300 internet services signifies the extensive reach of AI and deep learning.

Transcript

in France is the last stage of the machine learning pipeline this is where you take the model that you train and deploy it into services to make predictions what comes out of the machine learning pipeline and frameworks are computational graphs that are incredibly complex these are gigantic computational graphs and there are so many different types... Read More

Questions & Answers

Q: What is TensorRT, and how does it help in the deployment of machine learning models?

TensorRT is an optimizing compiler developed by NVIDIA, used to deploy machine learning models by compiling the complex computational graphs into target machines for efficient execution. It supports various neural network architectures and precisions, making it easier to run models faster in real-world applications.

Q: How does Jarvis simplify the development of conversational AI services?

Jarvis is an application framework introduced by NVIDIA, aimed at simplifying the creation of conversational AI services. It includes pre-trained, state-of-the-art models that have been optimized for interactive performance. Additionally, it provides the Nemo tool, which allows developers to train the Jarvis models with their domain-specific data.

Q: What is the significance of Jarvis' ability to create interactive 3D chatbots?

With the help of technologies like audio-to-face models and computer graphics in Omniverse, Jarvis enables the creation of interactive 3D chatbots. This breakthrough capability enhances the conversational AI experience, as the chatbots can animate facial expressions while responding to queries, resulting in more immersive and engaging interactions.

Q: How does Jarvis contribute to real-time inference applications like video conferencing?

Jarvis plays a vital role in automating conversations and deploying conversational AI services in various new applications, including video conferencing. With Jarvis, it is possible to pick up multiple people's speech simultaneously, provide closed captions or real-time translations, and generate summaries and transcriptions at the end of a conference. This transforms the video conferencing experience and improves efficiency.

Summary & Key Takeaways

• NVIDIA has developed an optimizing compiler called TensorRT, which can handle various neural network architectures and supports different precisions, leading to a 10x increase in developers using it.

• NVIDIA has created an application framework named Jarvis, which enables the development of interactive 3D chatbots and simplifies the creation of conversational AI services.

• Jarvis utilizes models that have been pre-trained with a significant amount of data and computation, and it includes a tool called Nemo for augmenting the models with domain-specific data.