Zero-Shot Learning and Clustering for Semantic Utterance Classification
Authors: Yann N. Dauphin; Gokhan Tur; Dilek Hakkani-Tur; Larry Heck
ICLR 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of the zero-shot semantic learning algorithm on the SUC dataset collected by (Tur et al., 2012). Furthermore, we achieve state-of-the-art results by combining the semantic features with a supervised method. |
| Researcher Affiliation | Collaboration | Yann N. Dauphin1 Gokhan Tur2 Dilek Hakkani-T ur2 Larry Heck2 1University of Montreal, Montreal, Canada 2Microsoft Research, Mountain View, CA, USA |
| Pseudocode | No | The paper describes algorithms and models in text and mathematical formulas but does not include explicit pseudocode blocks or sections labeled "Algorithm". |
| Open Source Code | No | The paper does not contain any statements or links indicating that its source code is publicly available. |
| Open Datasets | No | We have gathered a month of query click log data from Bing to learn the embeddings. ... We evaluate the performance of the methods for SUC on the dataset gathered by (Tur et al., 2012). The paper mentions using specific datasets but does not provide concrete access information (e.g., links, DOIs, repositories, or explicit statements of public availability) for them. |
| Dataset Splits | Yes | There are 16,000 training utterances, 2000 utterances for validation and 2000 utterances for testing. |
| Hardware Specification | Yes | The models are trained on a cluster of computers with double quad-core Intel(R) Xeon(R) CPUs with 2.33GHz and 8Gb of RAM. |
| Software Dependencies | No | The paper mentions training deep neural networks and using SVMs but does not specify any software libraries or packages with their version numbers. |
| Experiment Setup | Yes | The learning rate parameter of gradient descent is found by grid search with {0.1, 0.01, 0.001}. The number of layers is between 1 and 3. The number of hidden units is kept constant through layers and is found by sampling a random number from 300 to 800 units. ... We sample the dropout rate randomly between 0% dropout and 20%. The λ of the zero-shot embedding method is found through grid-search with {0.1, 0.01, 0.001}. |