Online Neural Sequence Detection with Hierarchical Dirichlet Point Process
Authors: Weihan Li, Yu Qi, Gang Pan
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We manifest these advantages on synthetic data and neural recordings from songbird higher vocal center and rodent hippocampus. |
| Researcher Affiliation | Academia | 1College of Computer Science and Technology, Zhejiang University, Hangzhou, China 2The Affiliated Mental Health Center & Hangzhou Seventh People s Hospital, the MOE Frontier Science Center for Brain Science and Brain-machine Integration, Zhejiang University School of Medicine, Hangzhou, China 3The State Key Lab of Brain-Machine Intelligence, Zhejiang University, Hangzhou, China |
| Pseudocode | Yes | Algorithm 1: Particle filter for HDPP |
| Open Source Code | Yes | Our code is available at https://github.com/ Weihan Likk/Hierarchical-Dirichlet-Point-Process. |
| Open Datasets | Yes | We evaluate the proposed model on synthetic data as well as real-world data from songbird higher vocal center [15]2 and rodent hippocampus [3, 7, 8]4. 2http://github.com/Fee Lab/seq NMF 4http://crcns.org/data-sets/hc/hc-11 |
| Dataset Splits | Yes | To have a good knowledge of hyperparameter settings, we performed cross-validation via a "speckled holdout pattern" strategy, which was previously used in PCA [33], and has recently been applied to neural data [28, 29]. We then evaluated the performance of HDPP by calculating log-likelihood on the held-out set. |
| Hardware Specification | Yes | The test of performance was carried by a desktop with AMD Ryzen 7 5800X 8-Core Processor and 32 GB RAM. |
| Software Dependencies | No | The paper provides a link to its code repository but does not explicitly list the specific versions of software dependencies (e.g., Python, PyTorch, or other libraries) used for its implementation or experiments within the text. |
| Experiment Setup | Yes | For HDPP, we set the particle number to 20, and for PP-Seq, we adopt parallel MCMC to parallelize the computation as well as setting the number of Gibbs sweeps to 2100 [28]. For the sequence amplitude βk, we randomized the mean sequence amplitude a1/b1 and set the variance of sequence amplitude a1/b2 1 equals to the mean. For the background noise amplitude, we randomized the prior rate b0 in gamma distribution and set the prior shape a0 to 100. |