Benchmarking the next generation of never-ending learners

In the new technical blog post NEVIS\u201922: A Stream of 100 Tasks Sampled From 30 Years of Computer Vision Research, author Marc Ranzton aims to study the question of efficient knowledge transfer in a controlled and reproducible setting. The Never-Ending Visual Classification Stream (NEVIS\u201922), as the stream is called, is comprised of 106 tasks extracted from major computer vision conference proceedings over the past three decades. NEVIS\u201922 aims to create an open-source codebase and evaluate learning models using a broad selection of algorithms. The challenges posed by the new stream include achieving both accurate and efficient knowledge transfer, as well as improving trade-offs between error rate and compute for learning future tasks. The paper includes data, methodology, and evaluation metrics to provide insight into the generalizability and effectiveness of NEVIS\u201922.