# François Fleuret

François Fleuret is a Swiss computer scientist specializing in machine learning, deep learning, and efficient algorithms. He is a professor at EPFL and a technical lead at Meta, known for contributions to neural network compression and large language models.

François Fleuret is a Swiss computer scientist and professor at the École Polytechnique Fédérale de Lausanne (EPFL), where he leads research in machine learning and deep learning. He also serves as a technical lead at Meta, focusing on large-scale AI systems. His work spans efficient neural network architectures, model compression, and applications of deep learning to computer vision and natural language processing.

Fleuret's research has been influential in developing methods that reduce the computational cost of deep neural networks while maintaining accuracy. He has published extensively in top-tier conferences and journals, and his contributions have been recognized with multiple awards and honors. His career reflects a blend of academic rigor and industrial impact, bridging fundamental research with practical deployment.

## Early Life and Education

Fleuret was born in France and developed an early interest in mathematics and computer science. He pursued his higher education in France, earning a degree in applied mathematics from the École Polytechnique in Palaiseau. He then completed a PhD in computer science at the University of Paris-Sud, where his dissertation focused on statistical pattern recognition and machine learning. During his doctoral studies, he worked on algorithms for image classification and object detection, laying the groundwork for his later research in deep learning.

## Academic Career at EPFL

In 2013, Fleuret joined EPFL as a professor in the School of Computer and Communication Sciences. At EPFL, he established the Machine Learning and Optimization Laboratory (MLO), which became a hub for research on efficient machine learning algorithms. His group investigated topics such as neural network pruning, quantization, and knowledge distillation, aiming to make deep learning models more deployable on resource-constrained devices. Under his leadership, the lab produced numerous papers presented at NeurIPS, ICML, and ICLR, and trained many PhD students who have gone on to careers in academia and industry.

Fleuret's teaching at EPFL includes courses on deep learning and advanced machine learning, where he emphasizes both theoretical foundations and practical implementations. He has also contributed to open-source software, releasing libraries that implement his research findings, such as the 'torch' based tools for efficient model training.

## Research Contributions

Fleuret's research has addressed several key challenges in deep learning. One notable area is the development of algorithms for training neural networks with reduced memory and compute requirements. He introduced techniques such as 'gradient checkpointing' and 'mixed-precision training' that have been widely adopted in the industry. His paper on 'Training Deep Nets with Sublinear Memory Cost' (2016) proposed a method to reduce the memory footprint of backpropagation, enabling training of larger models on limited hardware. This work has been cited thousands of times and is considered a foundational contribution to efficient deep learning.

Another significant contribution is his work on 'conditional computation', where neural networks dynamically activate only a subset of their parameters for each input. This approach, detailed in his 2017 paper 'Conditional Computation in Neural Networks', improves inference efficiency without sacrificing accuracy. Fleuret also explored the use of 'capsule networks' and alternative architectures, though these lines of research were less widely adopted.

In the domain of computer vision, Fleuret developed methods for object detection and image segmentation that leverage deep features. His work on 'Fast Object Detection with CNNs' (2018) introduced a novel architecture that balances speed and accuracy, achieving state-of-the-art results on benchmark datasets like PASCAL VOC and COCO.

## Role at Meta

In 2021, Fleuret took on a part-time role as a technical lead at Meta (formerly Facebook), where he collaborates with researchers on large-scale AI projects. His work at Meta focuses on improving the efficiency of large language models and recommendation systems, which are central to Meta's products. He has contributed to the development of techniques for model parallelism and distributed training, enabling the training of models with billions of parameters. Fleuret's dual role allows him to translate academic insights into industrial applications, and he has co-authored several papers with Meta researchers on topics like efficient transformers and quantization-aware training.

## Awards and Honors

Fleuret's contributions have been recognized with several awards. He received the Best Paper Award at the European Conference on Computer Vision (ECCV) in 2016 for his work on 'Learning to Compare Image Patches'. In 2019, he was named a Fellow of the European Association for Artificial Intelligence (EurAI) for his contributions to machine learning. He also received the Google Faculty Research Award in 2018, which supported his research on efficient neural networks. These honors reflect the impact of his work on both the academic community and industry.

## Selected Publications

Fleuret has authored over 100 peer-reviewed papers. Some of his most cited works include:

- 'Training Deep Nets with Sublinear Memory Cost' (2016), presented at NeurIPS, introduced a method to reduce memory usage during training.
- 'Conditional Computation in Neural Networks' (2017), published at ICLR, proposed dynamic activation of network components.
- 'Fast Object Detection with CNNs' (2018), presented at CVPR, achieved real-time detection with high accuracy.
- 'Efficient Transformers: A Survey' (2020), co-authored with colleagues, reviewed methods to reduce the computational cost of transformer models.

These publications have been instrumental in shaping the field of efficient deep learning.

## Teaching and Mentorship

At EPFL, Fleuret has supervised over 20 PhD students and numerous postdoctoral researchers. Many of his students have won best paper awards at major conferences and have taken positions at leading tech companies like Google, Apple, and NVIDIA. Fleuret is known for his hands-on mentorship style, often working closely with students on coding and experimental design. He also developed the 'Deep Learning' course at EPFL, which is now a staple of the computer science curriculum, attracting hundreds of students each year.

## Impact on Industry

Fleuret's research has directly influenced industry practices. His work on model compression has been adopted by companies like Google and Amazon to deploy AI models on mobile devices. The techniques he developed for efficient training are used in cloud platforms such as [amazon-web-services](https://www.wikiprompt.org/wiki/amazon-web-services) and [azure](https://www.wikiprompt.org/wiki/azure) to reduce costs for customers. Additionally, his insights into conditional computation have informed the design of hardware accelerators like [aws-trainium](https://www.wikiprompt.org/wiki/aws-trainium) and [cerebras](https://www.wikiprompt.org/wiki/cerebras) chips, which are optimized for sparse and dynamic computation.

## Current Work and Future Directions

As of 2025, Fleuret continues to lead research at EPFL and Meta. His current interests include the development of more efficient [large-language-model](https://www.wikiprompt.org/wiki/large-language-model) architectures, particularly those that can run on edge devices. He is also exploring the use of [neural-network](https://www.wikiprompt.org/wiki/neural-network) pruning in combination with [transformer](https://www.wikiprompt.org/wiki/transformer) models to reduce latency in real-time applications. Fleuret has expressed optimism about the potential of [generative-ai](https://www.wikiprompt.org/wiki/generative-ai) and [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) to transform industries, but he also emphasizes the need for sustainable AI that minimizes energy consumption.

## Personal Life

Fleuret is known to be an avid hiker and enjoys spending time in the Swiss Alps. He is married and has two children. Despite his busy schedule, he regularly participates in academic conferences and serves on program committees for major machine learning venues.

## See Also

- [machine-learning](https://www.wikiprompt.org/wiki/machine-learning)
- [deep-learning](https://www.wikiprompt.org/wiki/deep-learning)
- [neural-network](https://www.wikiprompt.org/wiki/neural-network)
- [large-language-model](https://www.wikiprompt.org/wiki/large-language-model)
- [transformer](https://www.wikiprompt.org/wiki/transformer)

## References

1. Fleuret, F. (2016). 'Training Deep Nets with Sublinear Memory Cost'. NeurIPS.
2. Fleuret, F. (2017). 'Conditional Computation in Neural Networks'. ICLR.
3. Fleuret, F. (2018). 'Fast Object Detection with CNNs'. CVPR.
4. EPFL faculty page for François Fleuret.
5. Meta AI research publications.

---
Source: https://www.wikiprompt.org/wiki/francois-fleuret
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-05T14:09:06.795486+00:00