This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy

Decoded: A young researcher explains how AI works, and why we don't walk through red lights

(July 2024) Meet Etienne Filhol, a young French Huawei engineer who conducts research on artificial intelligence at the company’s “Paris Distributed and Parallel Technology Lab.” 

The Paris Distributed and Parallel Technology Lab focuses on new methods for automatic distribution of computation and data on large parallel heterogeneous machines. A particular focus is given to AI codes, for training and inference alike.

Etienne Filhol (right), a young AI engineer who goes places, is Zach's guest in this instalment of Decoded. 

Etienne joined Huawei in 2021 following an internship at the company the previous year. He was soon offered opportunities that few organizations can match. For instance, in early 2022, he was given the chance to work for six months at the company’s majestic Sanyapo campus in Dongguan in southern China where some of the best software engineers in the country are based. 
At Huawei, Etienne works mostly on MindSpore. This open source AI framework has capabilities comparable to those of Facebook’s PyTorch and Google’s TensorFlow, but it requires about 20% fewer lines of codes.

Etienne had the chance to work at Huawei's Sanyapo research campus soon after he joined the company

MindSpore can deliver AI developers superior performance. Independent programmers and engineers have reported that the combination of MindSpore and systems built with Huawei's own AI chips provides double the performance of competing systems. 
Huawei launched MindSpore in Sept. 2019 and opened sourced it in March 2020. The framework offers interactive debugging tools, comprehensive graph visualization, and robust logging and profiling features, enabling developers to effectively trace errors, inspect computational graphs and monitor performance. These capabilities help pinpoint issues quickly, ensure code efficiency and optimize execution.
MindSpore also supports importing and exporting neural networks to and from other frameworks, enabling seamless interoperability. This feature lets developers leverage models and tools from other frameworks, facilitating integration and collaboration across AI ecosystems.
The global MindSpore community implements several aspects of open governance.  All community meetings are recorded and posted on Bilibili. And it’s possible to initiate Special Interest Groups without approvals from Huawei.  In addition, MindSpore offers transparency on whether developers’ patches are accepted.  Developers have visibility to the patch pipeline and where their proposed patch(es) sit in it. 

The Distributed and Parallel Technology Lab is located in Paris, France. 

Initially called Ask Huawei Anything during early development, Huawei Decoded is an entertaining series in which Huawei experts candidly answer top Internet questions. Previous episodes: 




  

More Videos