AnchorInv: Few-Shot Class-Incremental Learning of Physiological Signals via Feature Space-Guided Inversion
Authors: Chenqi Li, Boyan Gao, Gabriel Davis Jones, Timothy Denison, Tingting Zhu
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | When evaluated on three public physiological time series datasets, Anchor Inv exhibits efficient knowledge forgetting prevention and improved adaptation to novel classes, surpassing state-of-the-art baselines. We comprehensively benchmark our proposed method against SOTA methods on three physiological datasets, surpassing state-of-the-art baselines. |
| Researcher Affiliation | Academia | 1Department of Engineering Science, University of Oxford, United Kingdom 2Oxford Digital Health Labs, Nuffield Department of Women s & Reproductive Health, University of Oxford, United Kingdom EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1: Pseudocode for Anchor Inv |
| Open Source Code | Yes | Code https://github.com/chenqi-li/anchorinv |
| Open Datasets | Yes | Brain-Computer Interface Motor-Imagery The BCI-IV 2a dataset (Brunner et al. 2008) is an EEG dataset for the classification of 4 distinct motor imagery classes. Grading of Neonatal Hypoxic-Ischemic Encephalopathy The NHIE dataset (O toole et al. 2023) is an EEG dataset for grading the severity of background abnormalities in neonates who have been diagnosed with hypoxic-ischaemic encephalopathy, a common brain injury as a result of impaired oxygen or blood flow to the brain during the time of birth. Gesture Recognition The GRABMyo dataset (Jiang, Pradhan, and He 2022) is an EMG dataset for classifying 16 distinct hand and finger gestures. |
| Dataset Splits | Yes | We select the first two classes as base classes (left hand, right hand), and the two incremental classes (feet, tongue) are learned over two incremental sessions, with a 1-way-10-shot setup. We select the first 2 classes as base classes (normal/mildly abnormal, moderate abnormalities), and the remaining 2 incremental classes (major abnormalities, inactive) are learned over 2 incremental sessions, with a 1-way-10-shot setup. We select the first 10 classes as base classes, and learn 6 incremental classes over 6 incremental sessions, with a 1-way-10 shot setup. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. It only describes the model architecture and its modifications for different datasets. |
| Software Dependencies | No | The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, PyTorch 1.9, CUDA 11.1) needed to replicate the experiment. |
| Experiment Setup | Yes | For the classifier, gϕ(t), we follow the classifier described in Equations 3 to 5 with T = 16. Optimization can be performed via gradient descent on randomly initialized replay samples. |