Scalable Attentive Sentence Pair Modeling via Distilled Sentence Embedding

Authors: Oren Barkan, Noam Razin, Itzik Malkiel, Ori Katz, Avi Caciularu, Noam Koenigstein3235-3242

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate the effectiveness of DSE on five GLUE sentence-pair tasks. DSE significantly outperforms several ELMO variants and other sentence embedding methods, while accelerating computation of the query-candidate sentence-pairs similarities by several orders of magnitude, with an average relative degradation of 4.6% compared to BERT. Furthermore, we show that DSE produces sentence embeddings that reach state-of-the-art performance on universal sentence representation benchmarks. Our code is made publicly available at https://github.com/microsoft/Distilled Sentence-Embedding.
Researcher Affiliation Collaboration Oren Barkan,*1 Noam Razin,*12 Itzik Malkiel,12 Ori Katz,13 Avi Caciularu,14 Noam Koenigstein12 1Microsoft, 2Tel Aviv University, 3Technion, 4Bar-Ilan University
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks, only descriptive text and a schematic illustration (Figure 1).
Open Source Code Yes Our code is made publicly available at https://github.com/microsoft/Distilled Sentence-Embedding.
Open Datasets Yes For sentence-pair tasks, our evaluation includes several datasets from the GLUE benchmark: MRPC (Dolan and Brockett, 2005), MNLI (Williams et al., 2018), QQP, QNLI (Wang et al., 2018), and STS-B (Cer et al., 2017). ... Following (Conneau et al. 2017), we opt for pre-training DSE on the All NLI (MNLI + SNLI) dataset.
Dataset Splits Yes We evaluate DSE on five sentence-pair tasks from the GLUE benchmark (Wang et al. 2018). ... The best model was selected based on the dev set.
Hardware Specification Yes We conducted two experiments on a single NVIDIA V100 32GB GPU using Py Torch.
Software Dependencies No The paper mentions 'Py Torch' but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes We used the Adam optimizer (Kingma and Ba 2014) with minibatch size of 32 and a learning rate of 2e-5, except for STS-B, where we used a learning rate of 1e-5. The models were trained for 8 epochs.