The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Supporting TensorFlow at Airbnb with Alfredo Luque - TWiML Talk #244
Mar 28, 2019
In this discussion, Alfredo Luque, a software engineer at Airbnb, dives into the challenges of transitioning AI projects from concepts to scalable systems. He shares insights on enhancing Airbnb's image categorization processes using TensorFlow, addressing technical hurdles with a database of half a billion images. Alfredo also introduces BigHead, a library that simplifies model building and integration with TensorFlow, highlighting its unique features like real-time execution and improved visualization tools. The conversation showcases the importance of effective machine learning tooling for streamlined workflows.
AI Snips
Chapters
Transcript
Episode notes
Image Categorization Project
- Airbnb needed to categorize listing images for better user experience.
- A model was trained to categorize half a billion images, reducing processing time from months to days.
Pre-processing Optimization
- Optimize pre-processing steps, especially when dealing with image data.
- This significantly improves inference speed, crucial for large-scale image processing.
Importance of Backfilling
- Backfilling is crucial for applying models to historical data, ensuring consistent results.
- At Airbnb, backfilling is essential for tasks like image categorization and generating embeddings for training data.