Vector Quantization for NN Compression with Julieta Martinez - #498
Analysis
This podcast episode of Practical AI features Julieta Martinez, a senior research scientist at Waabi, discussing her work on neural network compression. The conversation centers around her talk at the LatinX in AI workshop at CVPR, focusing on the commonalities between large-scale visual search and NN compression. The episode explores product quantization and its application in compressing neural networks. Additionally, it touches upon her paper on Deep Multi-Task Learning for joint localization, perception, and prediction, highlighting an architecture that optimizes computation reuse. The episode provides insights into cutting-edge research in AI, particularly in the areas of model compression and efficient computation.
Key Takeaways
- •Exploration of the commonalities between large-scale visual search and neural network compression.
- •Discussion of product quantization and its application in compressing neural networks.
- •Presentation of an architecture for Deep Multi-Task Learning that reuses computation for joint localization, perception, and prediction.
“What do Large-Scale Visual Search and Neural Network Compression have in Common”