Bidirectional (BiDi) Deep Learning
adjective
Definition: Describing a sequence-processing architecture that processes information in both directions (forward and backward), especially in models such as bidirectional RNNs, so that each element can be interpreted using both preceding and following context [Jurafsky, Martin 2026].
Example in context: “Both bidirectional LSTMs and Transformers assume that the sequence that is to be encoded is available in full, to be processed either forwards and backwards (BiLSTMs) or as a whole (Transformers).” [Madureira, Schlangen 2020]
Synonym: BiDi
Related terms: forward-backward processing, bidirectional RNN