January 2023 Summaries

The blog post details the use of Predibase, a low-code declarative machine learning platform, and Ludwig, an open-source project, to develop and deploy a machine learning model that predicts customer review ratings using a multi-modal dataset. The dataset used comprises over 20,000 anonymized reviews from the Women's E-commerce Clothing sector, featuring diverse data types such as categorical, numerical, and text features. The approach involves establishing baseline models with neural networks and tree-based models, iterating to improve model performance by incorporating unstructured text data, and eventually fine-tuning advanced models like BERT for enhanced results. Predibase streamlines the integration of structured and unstructured data, enabling easy model iteration and operationalization, including deployment through batch or real-time inference. This process highlights the platform's ability to handle complex data efficiently, making it suitable for businesses aiming to leverage machine learning for customer sentiment analysis and other predictive tasks.

Machine learning in 2022 saw significant advancements, particularly with generative AI tools like ChatGPT and DALL-E, and its application in drug discovery, such as Google's DeepMind's achievements. As machine learning transitions from research to practical applications, various trends are predicted for 2023 and beyond. These include the increased use of machine learning in scientific research, the adoption of multi-modal model architectures, and the emergence of large language model services akin to AWS. Additionally, deep learning is becoming more accessible to data scientists, and industry standards for data management are enabling scalable ML model deployment. MLOps is expanding to support pre-trained models, and the democratization of AI tools is facilitating broader adoption across corporate environments. Open-source foundation models could challenge Big Tech's dominance, while decreased corporate spending is prompting a focus on demonstrable ROI from ML tools. Ultimately, machine learning is expected to become ubiquitous in software products, mirroring the integration of networking in the 1990s. Predibase aims to simplify ML deployment with its low-code platform, making it easier for a broad range of users to develop and deploy ML models.