Company Description:

Are you interested in tackling one of the hardest, most important open problems in deep learning: making training more efficient and cost-effective?  MosaicML is a startup doing just that. Motivated by dramatic increases in model sizes and computing costs, we focus on developing fundamental innovations to improve the efficiency of training. We conduct the scientific research necessary to ensure that the capabilities unlocked by modern deep learning are accessible to everyone and sustainably scalable.

MosaicML is unique in our approach to this problem: we’re focused on every level of the stack, including the choice of hardware, the machine learning systems, the training algorithm, and the multi-run workflow by which scientists arrive at final models. We’re developing methods that alter the training algorithm (optimization procedure, model, and data) and training infrastructure to reduce resource usage and make the most of every cycle, watt, and dollar. This requires rigorous empirical research. Our researchers study how neural networks learn, develop methods to leverage these findings, and understand whether they work as intended. In short, research at MosaicML involves advancing the scientific frontier in service of making deep learning practical and efficient for everyone.

Job description:

You will work with one or more researchers on a project that will advance our existing projects toward making neural network training more efficient. This may include:

• Adapting, improving, and evaluating a method from the literature.
• Designing an entirely new method.
• Composing together multiple methods to create new recipes for efficient training.
• Scientifically investigating how neural networks learn in practice.
• Exploring new approaches for training neural networks.

Your qualifications and qualities:

•  Required:
  • Pursuing an undergraduate or graduate degree in computer science or related fields (electrical engineering, neuroscience, physics, math, etc.).
  • Some proficiency with the fundamentals of deep learning.
  • Proficient software engineering skills, including with PyTorch.
• Nice to have:
  
  • Knowledge of the systems aspects of how neural networks train and the resources utilized in the process of doing so.
  • Research experience in deep learning.

Additional Notes:

• US work authorization required
• Hourly Rate: $58.00