Efficient Anomaly Detection of Irregular Sequences in Ct-Echo Model Space
Authors: Ao Chen, Xiren Zhou, Huanhuan Chen
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This section details experiments tested on several datasets, followed by analysis of comparative experimental results and the presentation of additional analytical experiments. |
| Researcher Affiliation | Academia | Ao Chen, Xiren Zhou*, Huanhuan Chen* School of Computer Science and Technology, University of Science and Technology of China EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: Ct-Echo-Decay Input: Sequence x, sampling time t, number of sampling points E, input and hidden weights Whx and Whh, function fθ. Output: The output weights {Wyh, b} 1: Initialize h(t1) = tanh(Whx x(t1)) 2: for n = 2 to E do 3: hm(tn) = ODESolver(fθ(h), h(tn 1), (tn 1, tn)), 4: h(tn) = tanh(Whh hm(tn) + Whx x(tn)), 5: Compute weights of x(t + 1) = Wyhh(t) + b on observed timestep through Ridge Regression 6: return {Wyh, b} Algorithm 2: Ct-Echo-Leaky Input: Sequence x, sampling time t, number of sampling points E, input weight Whx, hidden weight Whh, function fθ. Output: The output weights {Wyh, b} 1: Initialize h(t0) = 0 2: x Cubic Spline Interpolation(x) 3: for n = 1 to E do 4: hm(tn) = ODESolver(fθ(h, x), h(tn 1), (tn 1, tn)), 5: h(tn) = tanh(hm(tn)), 6: Compute weights of x(t + 1) = Wyhh(t) + b on observed timestep through Ridge Regression 7: return {Wyh, b} |
| Open Source Code | No | The paper does not contain any explicit statements about code availability, nor does it provide a link to a code repository or supplementary materials containing code. |
| Open Datasets | Yes | Three datasets, CWRU, SU, and WHU are utilized, introduced as follows7. CWRU Dataset (Loparo 2012): Comprises 2-dimensional vibration signals from bearings under four load conditions (sub-datasets A, B, C, D), each with 200 instances per category. A mixed dataset is denoted as E. The dataset, with 2048 time steps per instance, supports a 10-category classification task (one normal category and nine distinct anomaly categories). SU Dataset (Shao et al. 2019): Contains 8-dimensional data for gearbox and bearing under two working conditions, denoted as sub-datasets G20, G30 for gearboxes, and B20, B30 for bearings, each with 1000 instances per category. The dataset includes four faults and one normal category, and each sample consists of 1024 time steps. WHU Dataset (Liu et al. 2019): Includes 1-dimensional vibration data from a rotor system, collected at 1200 revolutions per minute and a sampling frequency of 2048 Hz. Each instance is 2048 time steps long, with 45 instances per category. It encompasses four categories: normal, contact-rubbing, unbalanced, and misalignment. |
| Dataset Splits | Yes | for training, we use only 200 sequences per sub-dataset for CWRU and SU, and 90 for WHU; for testing, we use 1800 sequences per sub-dataset for CWRU, 4800 per sub-dataset for SU, and 90 for WHU. |
| Hardware Specification | Yes | The experiments are conducted on a PC with CPU: Intel(R) Core(TM) i9-13900k, and GPU: NVIDIA Ge Force RTX 3080. |
| Software Dependencies | No | The paper mentions several methods and techniques such as 'SVM', 'KNN', 'NFFT', 'GRU', 'Transformer', 'Runge-Kutta (4,5)', and 'Cubic Spline Interpolation' but does not specify any software libraries or frameworks with their version numbers that were used for implementation. |
| Experiment Setup | Yes | In our experiments, the spectral radius and size of the reservoir are set to 0.8 and 10 respectively. The decay rate of Proposed(Decay) is -0.05 and the leaky rate of Proposed(Leaky) is 1. |