Shot(f ,x,W,k ,{Mi}ki=1 ,H) 1 . Initializeα

Algorithms 1 and 2 respectively provide pseudo-code for theOne-Shot and Binary algorithms detailed in Section 3.3. This section provides further details and experiments for HopSupSup (introduced in Section 3.5). B.1 Training Recall that HopSupSup operates in ScenarioGNu and so task identity is known during training. The weights of the Hopfield network areΨ and µ stores a running mean of all masks learned so far. For a new taskk we use the same algorithm as in Section 4.2 to learn a binary maskmi which performs well fortaskk.