Mihajlo

Grbovic's research addresses the "cold-start" problem—recommending new items or assisting new users with no history—by using to explore heterogeneous data like text, images, and user behavior.

: His models capture both short-term intent (current search session) and long-term preferences (past bookings) to re-rank search results in milliseconds.

: Using sophisticated strategies to select "negative" examples (items a user saw but didn't click) to sharpen the model's ability to distinguish preferences. mihajlo

: Leveraging data from popular categories to improve recommendations in "thin" or niche markets. Seminal Works by Mihajlo Grbovic

In his papers, Grbovic outlines specific lessons for building production-grade deep learning systems: : Leveraging data from popular categories to improve

While Dr. Grbovic is the leading figure in deep learning, other notable "Mihajlos" contribute to various deep fields:

: He integrates visual data (photos) and textual metadata into a single hybrid model, ensuring that recommendations are not just based on clicks, but on the actual content of the listing. Key Technical Contributions Key Technical Contributions : He popularized applying the

: He popularized applying the "Word2vec" concept to marketplaces, treating a user's click-stream as a "sentence" and individual listings as "words" to learn high-quality embeddings.