Home  |  Publications  |  Talks  |  Service  |  Teaching  |  Honors  |  Google Scholar  |  GitHub  |  WeChat  |  Twitter

Talks & By-Invitation Events

This tutorial introduces programmable network devices and P4. The tutorial focuses on programming the P4 platform (v1model architecture and BMv2 Simple Switch target). Specifically, the platform is the Raspberry Pi powered by P4Pir.

This guest lecture is invited by Dr. Masoud Hemmatpour from UiT. The presentation will include the concept and benefits of in-network machine learning, algorithm mappings, hybrid and distributed deployment strategies, as well as several use cases.

This presentation will introduce the concept of in-network machine learning, the technology and its implementation, including three general model mapping methodologies. As in-network machine learning is a resource-constrained machine learning problem, two solutions will be introduced: a distributed deployment, and a hybrid deployment solution.

In this presentation, I presents DINC, a framework enabling distributed in-network computing, generating deployment strategies, overcoming resource constraints and providing functionality guarantees across a network.

This workshop brings together world-renowned experts in networking and AI to explore the exciting interplay of both to present and discuss cutting-edge developments... My presentation in this workshop mainly summarises two works, Planter and DINC, related to In-network machine learning.

In this presentation, I presents DINC, a framework enabling distributed in-network computing, generating deployment strategies, overcoming resource constraints and providing functionality guarantees across a network..

In this presentation, I presents DINC, a framework enabling distributed in-network computing, generating deployment strategies, overcoming resource constraints and providing functionality guarantees across a network..

Future of Internet Routing & Addressing (FIRA) includes works on use cases, design principles, architectures, and techniques, as well as insights into implementations and experiments with novel solutions that address the highlighted objectives. Our QCMP presentation focuses on designing in-network Q-learning for data centre load balancing.

Coseners is one of the important events for the UK networking society. This year, my presentation (a summary of my DPhil) includes all the work I (and the Computing Infrastructure Group, Oxford) have done related to In-network machine learning.

IoT devices are widely and dynamically deployed in diverse use cases, leading to a surge of security threats that were previously overlooked. We propose P4Pir, an in-network traffic analysis solution, to enable continuous learning and consistent ML updates. P4Pir achieves real-time multi-protocol data collection, in-network ML-based attack mitigation, and hitless runtime ML updates.

This presentation elucidates in-network machine learning methodologies, solutions of each model, the design of the Planter framework, the hybrid deployment technique, evaluations, as well as their applications.

This presentation will address real world challenges faced by in-network inference (limited resources, co-existing with core switch functionality, ML performance, retraining and others) and explain how we solved them.

In this presentation, I introduced our in-network machine learning inference research during the past several years, ranging from IIsy and Planter to P4Pir and Linnet.

In this presentation, I introduce in-network ML and its applications on smart IoT gateways and accelerating financial transactions. The evaluation shows that in-network ML have minor accuracy loss and can achieve near-optimal classification results, while significantly reducing latency and load on the servers.

This work designs Planter, a novel approach, which automatically and efficiently maps ML models to run within off-the-shelf network devices. My research will promote the practical application of network devices as ML accelerators to increase performance and reduce the load on computing infrastructure.

This work designs Planter, a modular framework to offload Machine Learning algorithms to programmable networked devices, which enable the internet to provide Machine Learning inference services on programmable network devices. This will revolutionise the way we use the internet for Machine Learning.

About 80 academic attendees and 80 Googler attendees. Our goal for this workshop is to expose academic researchers to the technical challenges that Google is facing, as well as some of the real-world constraints we operate under. Just as important, the workshop will give a chance to faculty in computer networking (and related areas) to expose their own new or early-stage work to each other, and to the Googlers who will be attending.

This work designs Planter, an algorithm for efficient mapping of machine learning models to programmable data planes. Planter increases the number of trees that can be mapped to a network device by overlapping trees within feature tables. And Planter can also support several different traditional models.

This work design an algorithm for efficient mapping of ensemble models, such as XGBoost and Random Forest, to programmable switches. The proposed method overlaps trees within match-action tables, achieves high accuracy and low resource overhead.