Рaycheck: 120-180к net for full-time

Format: part-time (20 hours/week), remote work, part-time job at MIPT

Bonus: publications in journals and at conferences based on project results

Detail:
At the Compression Group, we thrive on the breadth of AI. We use SotA methods of quantization, distillation and pruning of neural networks to reduce the memory footrpint of models and maintain their quality. We help our customers reduce infrastructure costs, speed up top grids and adapt them to low-power devices.

We invite experienced ML engineers who are interested in SotA methods of model compression and want to delve into this area: participate in R&D projects, create new methods and frameworks.

Tasks:

• Propose and test hypotheses on mesh compression with minimal loss of quality;
• Search and analyze articles on the topic of the project, discuss them with the team and customers;
• Implement SotA methods for compressing models and modify them to achieve the goals of the project;
• Conduct experiments, propose conclusions and ideas based on them for the next steps of the study;
• Log your work and share it with the team.

Knowledge and skills:

• You have experience with popular neural network architectures (ResNet, UNet, Fairseq, BERT, ViT etc);
• You understand neural network compression methods (Quantization-Aware Training, Post-Training Quantization, Knowledge Distillation, Structured/Unstructured Pruning etc), read and reproduced articles on these topics;
• You know how to apply compression methods in practice and work with specialized libraries (for example, torch.quantization);
• You like a scientific approach, you know how to generate and test hypotheses, to produce reproducible and reliable results;
• Make the most of PyTorch, Git, and your favorite IDE;
• You are able to work in a distributed team and use the best practices of collaborative development.

It will be an advantage if:

• Do you have scientific publications;
• You know how to identify customer needs and present a solution in terms of customer value.

Next steps:

• Leave an application on the site (P.S. include a portfolio and an open project in your resume);
• We will review the application and conduct an interview with the team;
• Perhaps we will offer you a small test task.