jfrankle.com - Jonathan Frankle - Chief Neural Network Scientist at Databricks MosaicML

Example domain paragraphs

I currently work as Chief Scientist at MosaicML , a startup dedicated to reducing the cost of training neural networks by changing the training algorithm itself. I lead the research team, which empirically studies the learning dynamics of practical neural networks, develops interventions that change the training algorithm to improve efficiency, and combines these speedup methods into recipes.

Our current recipes have reduced the cost and training time required to train ResNet-50 on ImageNet by 7x, DeepLabv3 on ADE-20K by 5x, BERT Pre-Training by 3x, and GPT Language Modeling by 2x. You can access these recipes through our Composer library for PyTorch. For the latest data, see the MosaicML Explorer . For details on our speedup methods, see our documentation .

In the fall of 2023, I will be joining the faculty at Harvard as a member of the Computer Science Department. I will be teaching courses on machine learning and deep learning, and I will continue my empirical research on the properties of practical deep neural networks. If you are interested in working with me, you should apply to the PhD program in computer science at Harvard.

Links to jfrankle.com (5)