QAID: Question Answering Inspired Few-shot Intent Detection

Authors: Asaf Yehudai, Matan Vetzler, Yosi Mass, Koren Lazar, Doron Cohen, Boaz Carmeli

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our results on three few-shot intent detection benchmarks achieve state-of-the-art performance.
Researcher Affiliation Collaboration IBM Israel Research Lab , Hebrew University of Jerusalem
Pseudocode No The paper describes the framework and training process in text and with equations, but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper links to external libraries and architectures used (e.g., Hugging Face transformers, ColBERT, SupContrast, Faiss) but does not provide a direct link to their own implementation of QAID or explicitly state that their code for this paper is released.
Open Datasets Yes We experiment with three widely studied few-shot intent detection datasets... For readily use: https://github.com/jianguoz/Few-Shot-Intent-Detection/tree/main/Datasets
Dataset Splits Yes Table 1: Data statistics of the three intent detection datasets from Dialo GLUE. It lists #Train, #Vaild, #Test counts for each dataset.
Hardware Specification Yes Our fine-tuning takes only ten minutes on one NVIDIA V100 GPU
Software Dependencies No The paper mentions software like 'Hugging Face transformers library', 'ColBERT architecture', and 'Faiss Index', but does not specify their version numbers or any other software dependencies with specific versions.
Experiment Setup Yes We train our encoder for 20 epochs with a batch size of 64, a learning rate of 1e 5, a temperature parameter τ of 0.07 and λ = 0.1... We train our model for 10 epochs... We set the batch size to 32... We set the temperature to 0.07. We also set λclass and λmlm to 0.1 and 0.05, respectively.