Structured Inference for Recurrent Hidden Semi-markov Model
Authors: Hao Liu, Lirong He, Haoli Bai, Bo Dai, Kun Bai, Zenglin Xu
IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the proposed model in a number of tasks, including speech modeling, automatic segmentation and labeling in behavior understanding, and sequential multi-objects recognition. Experimental results have demonstrated that our proposed model can achieve significant improvement over the state-of-the-art methods. |
| Researcher Affiliation | Collaboration | Hao Liu1,2, Lirong He1, Haoli Bai3, Bo Dai4, Kun Bai5 and Zenglin Xu1 1 SMILE Lab, University of Electronic Science and Technology of China, Chengdu, China 2 Yingcai Honors College, University of Electronic Science and Technology of China, Chengdu, China 3 Dept. of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong 4 College of Computing, Georgia Institute of Technology, Atlanta, GA USA 5 Mobile Internet Group, Tencent Inc., Shenzhen, China |
| Pseudocode | Yes | Algorithm 1 Strucutured Inference Algorithm for SSNN |
| Open Source Code | No | The paper implements the proposed model based on Theano and Block & Fuel, citing the Theano website (1http://deeplearning.net/software/theano/). However, there is no explicit statement about releasing the source code for their specific methodology or a link to their own repository. |
| Open Datasets | Yes | We evaluate the SSNN on several datasets across multiple scenarios. Specifically, we first evaluate its performance of finding complex structures and estimating data likelihood on a synthetic dataset and two speech datasets (TIMIT & Blizard). Then we test the SSNN for learning segmentation and latent labels on Human activity [Reyes-Ortiz et al., 2016] dataset, Drosophila [Kain et al., 2013] and Physio Net Challenge[Springer et al., 2016]. |
| Dataset Splits | Yes | For Blizzard we split the data using 90% for training, 5% for validation and 5% for testing. For TIMIT we use the predefined test set for testing and split the rest of the data into 95% for training and 5% for validation. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper states 'We implement the proposed model based on Theano1 and Block & Fuel [Van Merri enboer et al., 2015]. The Adam [Kingma and Ba, 2014] optimizer was used in all experiments for our model.' However, no specific version numbers are provided for Theano, Block & Fuel, or Adam. |
| Experiment Setup | Yes | the temperature τ starts from a large value 0.1 and gradually anneals to 0.01. In the experiment, we fix the τ at small value 0.0001, and the maximum time steps T 10, 000. In the experiment, we find that annealing of temperature τ is important, we start from τ = 0.15 and anneal it gradually to 0.0001. During training we use back-propagation through time (BPTT) for 1 second. We set the truncation of max possible duration M to be 400 for all tasks. We also set the number of hidden states K to be the same as the ground truth. |