Search for: All records

Due to the scarcity of reliable anomaly labels, recent anomaly detection methods leveraging noisy auto-generated labels either select clean samples or refurbish noisy labels. However, both approaches struggle due to the unique properties of anomalies.Sample selectionoften fails to separate sufficiently many clean anomaly samples from noisy ones, whilelabel refurbishmenterroneously refurbishesmarginalclean samples. To overcome these limitations, we design Unity, thefirstlearning from noisy labels (LNL) approach for anomaly detection that elegantly leverages the merits of both sample selection and label refurbishment to iteratively prepare a diverse clean sample set for network training. Unity uses a pair of deep anomaly networks to collaboratively select samples with clean labels based on prediction agreement, followed by a disagreement resolution mechanism to capture marginal samples with clean labels. Thereafter, Unity utilizes unique properties of anomalies to design an anomaly-centric contrastive learning strategy that accurately refurbishes the remaining noisy labels. The resulting set, composed ofselected and refurbishedclean samples, will be used to train the anomaly networks in the next training round. Our experimental study on 10 real-world benchmark datasets demonstrates that Unity consistently outperforms state-of-the-art LNL techniques by up to 0.31 in F-1 Score (0.52 \rightarrow 0.83). 

Background: The number of survivors of cancer is growing, and they often experience negative long-term behavioral outcomes due to cancer treatments. There is a need for better computational methods to handle and predict these outcomes so that physicians and health care providers can implement preventive treatments. Objective: This study aimed to create a new feature selection algorithm to improve the performance of machine learning classifiers to predict negative long-term behavioral outcomes in survivors of cancer. Methods: We devised a hybrid deep learning–based feature selection approach to support early detection of negative long-term behavioral outcomes in survivors of cancer. Within a data-driven, clinical domain–guided framework to select the best set of features among cancer treatments, chronic health conditions, and socioenvironmental factors, we developed a 2-stage feature selection algorithm, that is, a multimetric, majority-voting filter and a deep dropout neural network, to dynamically and automatically select the best set of features for each behavioral outcome. We also conducted an experimental case study on existing study data with 102 survivors of acute lymphoblastic leukemia (aged 15-39 years at evaluation and >5 years postcancer diagnosis) who were treated in a public hospital in Hong Kong. Finally, we designed and implemented radial charts to illustrate the significance of the selected features on each behavioral outcome to support clinical professionals’ future treatment and diagnoses. Results: In this pilot study, we demonstrated that our approach outperforms the traditional statistical and computation methods, including linear and nonlinear feature selectors, for the addressed top-priority behavioral outcomes. Our approach holistically has higher F1, precision, and recall scores compared to existing feature selection methods. The models in this study select several significant clinical and socioenvironmental variables as risk factors associated with the development of behavioral problems in young survivors of acute lymphoblastic leukemia. Conclusions: Our novel feature selection algorithm has the potential to improve machine learning classifiers’ capability to predict adverse long-term behavioral outcomes in survivors of cancer. 

Log anomaly detection, critical in identifying system failures and preempting security breaches, finds irregular patterns within large volumes of log data. Modern log anomaly detectors rely on training deep learning models on clean anomaly-free log data. However, such clean log data requires expensive and tedious human labeling. In this paper, we thus propose a robust log anomaly detection framework, PlutoNOSPACE, that automatically selects a clean representative sample subset of the polluted log sequence data to train a Transformer-based anomaly detection model. Pluto features three innovations. First, due to localized concentrations of anomalies inherent in the embedding space of log data, Pluto partitions the sequence embedding space generated by the model into regions that then allow it to identify and discard regions that are highly polluted by our pollution level estimation scheme, based on our pollution quantification via Gaussian mixture modeling. Second, for the remaining more slightly polluted regions, we select samples that maximally purify the eigenvector spectrum, which can be transformed into the NP-hard facility location problem; allowing us to leverage its greedy solution with a (1-(1/e)) approximation guarantee in optimality. Third, by iteratively alternating between the above subset selection, a model re-training on the latest subset, and a subset filtering using dynamic training artifacts generated by the latest model, the data selected is progressively refined. The final sample set is used to retrain the final anomaly detection model. Our experiments on four real-world log benchmark datasets demonstrate that by retaining 77.7\% (BGL) to 96.6\% (ThunderBird) of the normal sequences while effectively removing 90.3\% (BGL) to 100.0\% (ThunderBird, HDFS) of the anomalies, Pluto provides a significant absolute F-1 improvement up to 68.86\% (2.16\% → 71.02\%) compared to the state-of-the-art sample selection methods. The implementation of this work is available at https://github.com/LeiMa0324/Pluto-SIGMOD25.