January 23, 2024
We are delighted to announce that Blaise Delattre, a CIFRE thesis student with Foxstream, and Quentin Barthélemy, Machine Learning researcher, will be taking part in the prestigious ICLR 2024 conference in Vienna, Austria, from 7 to 11 May. They were selected from among 8,000 submissions for their paper on the robustness of artificial intelligence models.
The paper proposes an innovative approach based on injecting random noise into neural networks to stabilise their predictions. In this paper, our researchers establish new links between certain key properties of these models. Their work details a technique that significantly improves the robustness of neural networks.
The results show significant gains in terms of certified robustness. The certification approach can also be applied to reinforce models that have already been trained.
Thanks to these advances, we are actively exploring ways of making our solutions even more effective, thereby contributing to the ongoing development of our expertise in the field of artificial intelligence.