GRADES: Gradient Descent for Similarity Caching

A similarity cache can reply to a query for an object with similar objects stored locally. In some applications of caches, queries and are naturally represented as points in continuous space. This is example the case $360^\circ$ videos where user’s head orientation—expressed spherical coordinates—determines what part video needs be retrieved, or recommendation systems metric learning technique used embed finite dimensional space opportune distance capture content dissimilarity. Existing caching policies simple modifications classic like LRU, LFU, qLRU ignore nature embedded. this paper, we propose Grades, new policy that uses gradient descent navigate find appropriate store cache. We provide theoretical convergence guarantees show Grades increases served by both mentioned above.