IC Colloquium - Finally: Autonomous Systems
By: JP Vasseur - NVIDIA
Abstract
For two decades, the promise of truly autonomous systems—characterized by self-learning, self-healing, and self-governance—has remained a persistent, yet largely unrealized, ambition in computer science. We posit that this inflection point has now arrived. The synergistic convergence of deep statistical analysis, predictive machine learning, and sophisticated agentic frameworks powered by generative AI provides, for the first time, the technological foundation to achieve this long-sought autonomy.
This presentation introduces a novel architectural framework for a truly autonomous, self-improving system designed for the operational management of large-scale, complex distributed environments. We propose a cognitive system built upon a fleet of collaborative, specialized AI agents. This multi-agent architecture integrates heterogeneous models—spanning statistical machine learning for predictive analytics and generative AI for complex reasoning and knowledge synthesis. We will deconstruct the core principles of this approach, addressing key scientific challenges inherent in building such autonomous systems.
Core topics will include: managing system stochasticity; ensuring the interpretability of agentic decision-making; the efficacy of knowledge compression techniques for interfacing with Large Language Models (LLMs); and methodologies for measuring operational efficacy in a closed-loop system. We will demonstrate why a proactive, predictive posture is a critical for autonomy, exploring the role of machine learning in identifying failure patterns before they manifest. Furthermore, we will analyze the challenges of emergent reasoning trajectories within the multi-agent system and outline a framework for enabling continuous self-improvement. This work presents a conceptual and practical blueprint for the next frontier of autonomous systems, moving beyond mere automation to achieve genuine operational self-governance.
Bio
JP Vasseur is a pioneering technologist whose innovations have shaped Networking, Artificial Intelligence, and large-scale compute systems for over three decades. A key contributor to Networking( PCE, MPLS, and Traffic Engineering), he has also led advancements in Machine Learning and Generative AI, driving the convergence of AI and infrastructure in the Internet (WAN, Wifi, DC). As Senior Distinguished Engineer and Chief Architect, AI & Networking at NVIDIA, JP leads the design of agentic, self-operating data centers capable of autonomous issue detection, root-cause analysis, and remediation. Previously, during more than 25 years at Cisco, including 13 as Fellow and AI VP of Engineering, he pioneered AI-driven solutions and led team shipping products deployed at large scale. (in IoT, SD-WAN, Wifi and Security)
Holder of over 750 patents and a PhD in Computer Science, JP is a globally recognized inventor, author, and thought leader advancing the intersection of AI, networking, and neuroscience.
More information
Abstract
For two decades, the promise of truly autonomous systems—characterized by self-learning, self-healing, and self-governance—has remained a persistent, yet largely unrealized, ambition in computer science. We posit that this inflection point has now arrived. The synergistic convergence of deep statistical analysis, predictive machine learning, and sophisticated agentic frameworks powered by generative AI provides, for the first time, the technological foundation to achieve this long-sought autonomy.
This presentation introduces a novel architectural framework for a truly autonomous, self-improving system designed for the operational management of large-scale, complex distributed environments. We propose a cognitive system built upon a fleet of collaborative, specialized AI agents. This multi-agent architecture integrates heterogeneous models—spanning statistical machine learning for predictive analytics and generative AI for complex reasoning and knowledge synthesis. We will deconstruct the core principles of this approach, addressing key scientific challenges inherent in building such autonomous systems.
Core topics will include: managing system stochasticity; ensuring the interpretability of agentic decision-making; the efficacy of knowledge compression techniques for interfacing with Large Language Models (LLMs); and methodologies for measuring operational efficacy in a closed-loop system. We will demonstrate why a proactive, predictive posture is a critical for autonomy, exploring the role of machine learning in identifying failure patterns before they manifest. Furthermore, we will analyze the challenges of emergent reasoning trajectories within the multi-agent system and outline a framework for enabling continuous self-improvement. This work presents a conceptual and practical blueprint for the next frontier of autonomous systems, moving beyond mere automation to achieve genuine operational self-governance.
Bio
JP Vasseur is a pioneering technologist whose innovations have shaped Networking, Artificial Intelligence, and large-scale compute systems for over three decades. A key contributor to Networking( PCE, MPLS, and Traffic Engineering), he has also led advancements in Machine Learning and Generative AI, driving the convergence of AI and infrastructure in the Internet (WAN, Wifi, DC). As Senior Distinguished Engineer and Chief Architect, AI & Networking at NVIDIA, JP leads the design of agentic, self-operating data centers capable of autonomous issue detection, root-cause analysis, and remediation. Previously, during more than 25 years at Cisco, including 13 as Fellow and AI VP of Engineering, he pioneered AI-driven solutions and led team shipping products deployed at large scale. (in IoT, SD-WAN, Wifi and Security)
Holder of over 750 patents and a PhD in Computer Science, JP is a globally recognized inventor, author, and thought leader advancing the intersection of AI, networking, and neuroscience.
More information
Practical information
- General public
- Free
Contact
- Host: Martin Schrimpf