Large Language Model Guided Knowledge Distillation for Time Series Anomaly Detection
Authors: Chen Liu, Shibo He, Qihang Zhou, Shizhong Li, Wenchao Meng
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Comprehensive experiments are conducted to demonstrate the superiority of our method on 9 univariate datasets and 6 multivariate datasets. and 4 Experiment |
| Researcher Affiliation | Academia | Chen Liu1 , Shibo He1 , Qihang Zhou1 , Shizhong Li1 and Wenchao Meng1 1Zhejiang University {liu777ch, s18he, zqhang, lisz, wmengzju}@zju.edu.cn |
| Pseudocode | No | The paper describes the methodology using prose and a framework diagram (Figure 2), but it does not include structured pseudocode or an algorithm block. |
| Open Source Code | No | The paper does not provide an explicit statement or link to open-source code for the described methodology. |
| Open Datasets | Yes | UCR Anomaly Archive (UCR). This archive comprises 250 diverse univariate time series signals spanning various domains [Wu and Keogh, 2021]. and We also evaluate our method on 6 previously commonly used multivariate datasets, including SMD, MSL, SMAP, PSM, Wa Q, and SWAN. |
| Dataset Splits | No | Tables 1 and 2 list "Train" and "Test" data, but there is no explicit mention or details of a "validation" split in the text or tables. |
| Hardware Specification | Yes | All experiments are conducted on a single RTX 3090. |
| Software Dependencies | No | The paper mentions software components like "Adam optimizer" and "GPT2" but does not specify their version numbers or other ancillary software dependencies. |
| Experiment Setup | Yes | During the training stage, we utilize an Adam optimizer with a learning rate of 0.0001 and a batch size of 32. All experiments are conducted on a single RTX 3090. and For our model, we use the pretrained GPT2 with 6 layers as our teacher network. Regarding the student network, we use a pool with 32 prototypes and an attention mechanism with an intermediate dimension of 64 and a head number of 8. |