Hogwild is scheme developed to run in parallel with the Stochastic gradient descent (SGD). It is a strategy to eliminate overhead correlated with synchronization and locking when running with SGD.
In Hogwild, the processors are allowed equal access to shared memory and can update individual components of memory . Threads are allowed to overwrite one another and gradients are computed on stale versions of the current solution.
Timeline
No Timeline data yet.
