Identifying and Controlling Important Neurons in Neural Machine Translation

Authors: Anthony Bau, Yonatan Belinkov, Hassan Sajjad, Nadir Durrani, Fahim Dalvi, James Glass

ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show experimentally that translation quality depends on the discovered neurons, and find that many of them capture common linguistic phenomena.
Researcher Affiliation Collaboration 1MIT Computer Science and Artificial Intelligence Laboratory, Cambridge, MA 02139, USA 2Qatar Computing Research Institute, HBKU Research Complex, Doha 5825, Qatar
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is publicly available as part of the Neuro X toolkit (Dalvi et al., 2019b).7https://github.com/fdalvi/Neuro X
Open Datasets Yes We use the United Nations (UN) parallel corpus (Ziemski et al., 2016) for all experiments.
Dataset Splits No The paper mentions training models on "different parts of the training set" and evaluating on "the official test set", but does not explicitly detail a separate validation set split.
Hardware Specification No The paper does not provide specific details regarding the hardware (e.g., GPU models, CPU types) used for running the experiments.
Software Dependencies No The paper mentions using Spacy for linguistic annotations and refers to a character convolutional neural network (char CNN) and LSTM encoder-decoder models, but it does not specify version numbers for any software dependencies or libraries.
Experiment Setup Yes We train 500 dimensional 2-layer LSTM encoder-decoder models with attention (Bahdanau et al., 2014). In order to study both word and sub-word properties, we use a word representation based on a character convolutional neural network (char CNN) as input to both encoder and decoder.