Machine Learning Assessment Question Which deep learning architecture is suitable for...
Fujitsu technical mcq question, verified with a worked answer. Free to practise - no sign-up.
Machine Learning Assessment Question Which deep learning architecture is suitable for sequence-to-sequence tasks like machine © Convolution Neural Network (CNN) translation? - A) Recurrent Neural Network (RNN) - B) Gated Recurrent Unit (GRU) - C) Transformer
Show answer & explanation
While RNNs and GRUs can handle sequences, the Transformer architecture is the current state-of-the-art for sequence-to-sequence tasks like machine translation due to its self-attention mechanism, which allows for global dependencies and parallel processing.
Step-by-step Derivation:
Step 1: Analyze the requirement 'sequence-to-sequence tasks like machine translation'. This requires an architecture that can map an input sequence of variable length to an output sequence of variable length.
Step 2: Evaluate Option A (RNN). RNNs process sequences sequentially and suffer from vanishing/exploding gradients, making them poor for long-range dependencies in translation.
Step 3: Evaluate Option B (GRU). GRUs are a gated version of RNNs that mitigate the vanishing gradient problem but still process data sequentially, limiting training speed and long-term memory efficiency.
Step 4: Evaluate Option C (Transformer). Transformers replace recurrence with 'Self-Attention', allowing the model to weigh the importance of every word in a sentence regardless of position. This enables massive parallelism and superior performance in translation (e.g., Google Translate).
Step 5: Compare the three. While A and B are technically capable of seq2seq, the Transformer is the definitive 'suitable' modern architecture that defines the current paradigm of machine translation.