Using machine learning, NLP, and deep domain knowledge, Auchan Retail International achieved an impressive 18% reduction in out-of-stock items and overstock across national operations in just one year. Their dual-model strategy, extensive feature engineering, and close collaboration with stakeholders led to substantial operational improvements and efficiency in retail forecasting.

The paper discusses the superiority of Kalman Filter (KF) over neural networks in some cases and the need to optimize KF parameters. Despite its 60-year-old linear architecture, the KF outperformed a fancy neural network after parameter optimization. The study emphasizes the importance of optimizing KF and not relying on its assumptions, offering a simple training…

The article emphasizes the shift from creating traditional dashboards to storytelling with data, highlighting the need for more engaging and impactful communication of insights. It stresses the importance of framing questions, collecting relevant data, and structuring the data story in various engaging formats. The piece concludes with a call to embrace data storytelling for better…

Google and MIT researchers propose SynCLR, a novel AI approach for visual representation learning using synthetic images and captions. The method leverages generative models to synthesize large-scale training data, demonstrating superior performance to existing methods. The team highlights potential improvements and invites further research. For more details, refer to the original Paper and Github.

Vald is a cloud-native, open-source distributed vector search engine addressing challenges in large-scale similarity searches. Its features include distributed indexing, auto-indexing with backups, custom filtering, and horizontal scaling, making it resilient and versatile. Vald offers lightning-fast search on billions of vectorized data points, supporting multiple languages through gRPC. It’s a vital tool for advanced unstructured…

The text discusses techniques to improve the efficiency of large language models (LLMs) through prompt compression, focusing on methods such as AutoCompressors and LongLLMLingua. The goal is to reduce inference costs and enable faster and accurate responses. The article compares different compression methods and concludes that LongLLMLingua shows promise for prompt compression in applications like…

AutoRT, SARA-RT, and RT-Trajectory expand on our previous Robotics Transformers to improve robots’ decision-making speed, understanding, and navigation in diverse environments.