ML Resources — July 27
Grid computing, a paradigm where independent entities can pool their resources to run high-computation workloads, has been successfully used for several scientific projects. This paradigm hasn’t been able to make its mark with machine learning projects yet, due to the computational disparities of the participating devices. Diskin et al. propose a solution called DeLOC (Distributed Deep Learning in Open Collaborations) that maximizes training throughput for the hardware available.
Jeremy Howard is one of the most acclaimed machine learning practitioners of our times. This is an inspiring talk about his journey into deep learning and its incredible effectiveness.
Recent years have seen a renewed focus on tackling racism in society, both in the society at large and in the field of research. Field et al. conduct a survey of 79 papers from the anthology that mention race, and show us how current works operationalize race in a way that can propagate existing hierarchies, and the different ways in which current data and models encode racial bias.
Graham Neubig gave a fascinating talk at the Lisbon Machine Learning Summer school about prompting and calibration of large language models. More videos from the summer school have been made available available here, including reinforcement learning and causality
Current transformer models condition on relatively longer contexts, but what part of these contexts matter the most to performance? O’Connor et al. find that models are relatively robust to shuffling word order and removing all words except nouns.
Aggregate Intellect
Aggregate Intellect is a Global Marketplace where ML Developers Connect, Collaborate, and Build. Connect with peers & experts at https://ai.science or Join our Slack Community.
- Check out the user generated Recipes that provide step by step, and bite sized guides on how to do various tasks
- Join our ML Product Challenges to build AI-based products for a chance to win cash prizes
- Connect with peers & experts through the ML Discussion Groups or Expert Office Hours