February 2023 Summaries

The State of Machine Learning Data Quality Report highlights the challenges faced by data scientists in fixing data errors, with 80% of their time spent on this task. The report explores issues across various data modalities and discusses how technology can help mitigate these problems. Galileo, a machine learning data quality intelligence tool, aims to inspect and fix data quality errors throughout the ML process, built on the experience of former Uber and Google AI engineers. By focusing on high-quality data, Galileo seeks to create a more productive and bias-free future for the world, allowing data scientists to save time and focus on more enjoyable tasks.

Debugging machine learning data can be a challenging task, but Galileo is designed to help developers debug their data 10x faster. With Galileo, you can inspect and fix data quality errors in all stages of the ML process, including natural language processing, computer vision, and structured data. The tool was developed by a team with first-hand experience working on large-scale machine learning projects at Uber and Google AI, who recognized the importance of high-quality data in creating a better future. Galileo aims to save developers time and focus on more enjoyable tasks, while also promoting bias-free outcomes. By attending a workshop or joining the free community offering, participants can gain a comprehensive understanding of the process of debugging machine learning data and learn how to tackle common challenges.

The goal of NER, or named entity recognition, is to identify and categorize entities in unstructured data, such as text or speech. In machine learning, Hugging Face has become a leading hub for pre-trained models and datasets, especially in Natural Language Processing (NLP) tasks like NER. Galileo is a tool that provides data quality analysis for Hugging Face's NLP pipelines, helping to identify mistakes and problems with the quality of training data. To use Galileo, users need to install the necessary libraries and create a new project and run. The tool provides an easy-to-use interface for uploading datasets, tokenizing text, aligning labels, and fine-tuning models like RoBERTa. After training, Galileo logs the current epoch and splits, providing insights into model performance and data quality. Users can view high-level summary statistics, detailed visualizations, and even specific data points to analyze their dataset's quality. By using Galileo, users can improve the accuracy of their NLP models and gain a better understanding of their training data.