Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore a 20-minute conference talk from the SNIA Compute+Memory+Storage Summit that delves into developing a standardized machine learning framework for semantic image retrieval within SSDs. Learn how to overcome the challenges posed by diverse analytics formats in deep learning approaches by implementing an advanced transformer model that converts various types of analytics into uniform embedding formats. Discover techniques for designing an embedded system that processes both analytics and user queries through SBERT transformer models, converting them into N-dimensional vectors for efficient storage and retrieval. Master the implementation of clustering techniques to create an intent-based interface that improves search accuracy and reduces false positives. Gain insights from Western Digital's Vishwas Saxena on building a unified framework that eliminates dependency on multiple database types and query languages, making image retrieval more efficient on low-compute storage devices.
Syllabus
Intro
Problems in Semantic Image Retrieval
Constraints
SSD with Unified Semantic Search Image Retrieval Fra
Solution
Siamese BERT - Sentence Transformer Model
Image Captioning
Design Decisions
Indexing of embedding using Annoy
Search on embedded systems
Application View - Add to Advanced Search
Application View - Semantic Image Retrieval
Taught by
SNIAVideo