Deep SVDD
A deep learning-based one-class classification method that trains a neural network to map normal samples near a center point in feature space and push anomalous samples away from the center. It is trained only on normal data and is suited to detecting unseen anomalies.
Timeline (last 90 days)
Under four disease scenarios, Deep SVDD achieves the highest average AUC across all tested finite CPU energy budgets.
On the NPU, Deep SVDD leads at a 1 mJ energy budget, but at 10 mJ and 100 mJ budgets, Monte Carlo Dropout has a slightly higher average performance.
The two-stage system (Deep SVDD + narrow CPU classifier) increases the proportion of inputs that are "correctly classified or correctly rejected" by 15.0–19.4 percentage points, a significant improvement over classifiers without a rejection mechanism.
At the retrospectively selected peak test accuracy for each method, the estimated energy consumption per input is 1.93–6.99 mJ for the Deep SVDD two-stage system, 52.53 mJ for Monte Carlo Dropout, and 8.76–17.51 mJ for deep ensembles.
Deep SVDD is a deep learning-based one-class classification method that is trained using only normal data, and at inference time a sample whose distance to the center exceeds a threshold is judged as anomalous.
All Facts (5)
Under four disease scenarios, Deep SVDD achieves the highest average AUC across all tested finite CPU energy budgets.
50%UnverifiedOn the NPU, Deep SVDD leads at a 1 mJ energy budget, but at 10 mJ and 100 mJ budgets, Monte Carlo Dropout has a slightly higher average performance.
50%UnverifiedThe two-stage system (Deep SVDD + narrow CPU classifier) increases the proportion of inputs that are "correctly classified or correctly rejected" by 15.0–19.4 percentage points, a significant improvement over classifiers without a rejection mechanism.
50%UnverifiedAt the retrospectively selected peak test accuracy for each method, the estimated energy consumption per input is 1.93–6.99 mJ for the Deep SVDD two-stage system, 52.53 mJ for Monte Carlo Dropout, and 8.76–17.51 mJ for deep ensembles.
50%UnverifiedDeep SVDD is a deep learning-based one-class classification method that is trained using only normal data, and at inference time a sample whose distance to the center exceeds a threshold is judged as anomalous.
50%