Trends in Deep Learning with Jeremy Howard - TWiML Talk #214

Dataset Access

Accessing NLP datasets poses significant challenges due to restrictive licensing agreements, hindering the ability to replicate research results. Jeremy Howard highlights the difficulties faced by students who cannot access datasets like the Reuters corpus without cumbersome procedures 1. This issue reflects a broader problem where datasets were created without considering the global community of researchers eager to contribute.

   

Model Sharing

Model sharing emerges as a promising solution in fields where data sharing is limited, such as medical imaging. Jeremy advocates for the release of pre-trained domain-specific networks, which can be fine-tuned and improved collaboratively across institutions 1. He explains that sharing model weights can yield results comparable to data sharing, fostering innovation without the need for direct data exchange.