Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on limited hardware resources and slowing down the inference time. To address this limitation, many techniques exist such as quantization and pruning. Developing models that are light and fast is a pre-requisite when deploying in real-life conditions, when speed is an important factor to make the authentication process as smooth and transparent for the end user as possible, and when large enough infrastructures may not be available.
Another challenge when running models in real life conditions is that models need to perform not only well but well on everyone, irrespective of their gender, origin, age, socio-economic situation etc. Conscious or unconscious bias is often cited as a shortcoming of facial verification systems. At Unissey, we are thriving for a universal use of technology and we are hence committed to provide algorithms that are actively fighting bias.
The internship objectives are hence twofold:
For both research topics, the intern will be asked to carry extensive research about the state-of-the-art techniques on the topic and to develop innovative solution using the existing literature and the intern’s knowledge of machine learning.
Details :
Quick screening call.
Technical test, to be completed at home.
Final interview, including a technical part, and a meeting with different members of the R&D team to evaluate your motivation / cultural fit.
These companies are also recruiting for the position of “Hardware Engineering”.