July 2023 Summaries

Cleanlab's open-source library is a popular software framework for practicing Data-Centric AI, which automatically detects various common issues in datasets such as label errors, outliers, near duplicates, and drift. The library can be used with almost any type of data, including images, text, tables, and audio. It provides flexible functionality that allows users to decide how to improve their dataset and model based on the outputs from its algorithms. Cleanlab Studio, a no-code platform, builds upon this foundation by automating most of the hard parts of turning raw data into reliable ML or Analytics, including labeling, training baseline models, diagnosing and correcting data issues, identifying the best ML model for your data, and deploying it to serve predictions in business applications. The platform offers an intuitive interface that allows users to employ just the right amount of automation to produce high-quality data quickly. Cleanlab software aims to automate much of the 80% of time spent discovering and correcting data issues in ML projects, enabling companies to achieve faster model deployment and business impact via ML on auto-corrected data. The future of AI is Data-Centric, with Cleanlab software playing a crucial role in improving reliability and efficiency in machine learning applications.

Large-scale datasets often contain errors that can lead to lower reliability and increased costs. Data-centric AI is a modern solution to this problem, but applying these techniques at scale was challenging until recently. Cleanlab Studio, a tool built on data-centric AI algorithms, can automatically analyze large datasets like ImageNet to find and fix issues such as mislabeled images, outliers, and near-duplicates. The tool also helps derive higher-level insights about the dataset as a whole, improving its quality and reliability for use in machine learning models and data analytics.

Cleanlab Studio automates the process of deploying machine learning (ML) models by detecting and correcting issues in the data, training a baseline model, identifying the best model for the dataset, retraining on the corrected data, and deploying it. The tool uses various AutoML systems and foundation models to learn about what doesn't look right in the dataset, and applies optimal combinations of large pretrained LLMs and fine-tuned Transformer networks for text datasets, CLIP/DINOv2 and fine-tuned computer vision networks for image datasets, and text models, neural architectures designed specifically for tabular data, and powerful tree ensembles like Gradient Boosting for tabular datasets. Users can quickly correct issues detected in their original dataset to improve its quality, retrain the model on the improved data, and deploy it with just a few clicks. Cleanlab Studio has been shown to outperform state-of-the-art models, including OpenAI Large Language Models, by improving the accuracy of deployed ML models, reducing errors by up to 28%, and making predictions quickly and at low costs. The tool is useful across many applications, beyond text datasets, and can handle arbitrary data types.

The Cleanlab team has successfully launched their flagship product, Cleanlab Studio for Enterprise, with $5 million in seed funding. The product aims to reduce 80-90% of the time and cost needed for teams to deploy data-driven AI solutions like LLMs, ML models, and business intelligence analytics that work more reliably and accurately on real-world datasets. By automating the process of finding and fixing outliers, label issues, and other data issues in image, text, and tabular datasets, Cleanlab Studio enables enterprise teams to train more reliable models and derive more accurate analytics and insights. The product has been pioneered at MIT, rooted in open-source, and proven by Fortune-500 companies such as BBVA and Google.

Cleanlab Studio is an AI platform that can automatically diagnose issues with synthetic data, such as identifying which synthetic examples do not look realistic, poorly represented modes/tails of the real data distribution, and high-fidelity resembling real data. The platform allows users to effortlessly perform text classification on their uploaded dataset without writing any code, providing a label issue score for each example based on the confidence of the model's predictions. This helps identify poor-quality synthetic data and other shortcomings of the synthetic dataset. By using Cleanlab Studio, users can automatically detect outliers in both real and synthetic data, pinpoint areas that need refinement in the synthetic data generation process, and ensure the quality and accuracy of their synthetic data to avoid unintended consequences such as overfitting, propagating biases, and handling domain gaps. The platform is particularly useful for text-based applications, but its principles apply broadly across other types of synthetic data, making it a valuable tool for addressing challenges related to data scarcity, privacy, and more in the machine learning landscape.

Cleanlab Studio is a no-code AI tool that helps E-commerce businesses improve their product listings, analytics, and customer experience by automatically identifying and correcting miscategorized products in large catalogs. By using Cleanlab Studio, retailers can boost SEO efforts, increase discoverability of products, and improve customer trust and confidence in the website. The tool provides an intuitive interface to visualize detected issues, correct miscategorized products, and gain valuable insights into data quality.