Backdoor attacks pose a serious and emerging threat to deep neural networks. By injecting trigger-labeled samples into the training data, these attacks cause compromised models to misclassify any input containing the trigger. Existing defenses typically rely on either recovering the trigger to remove the corresponding backdoor or performing additional fine-tuning to erase the backdoor functionality while retaining the legitimate functionality. Both strategies require access to a trusted clean dataset, an assumption that is often unrealistic in practice. We introduce Double-Blind Cleanser (DBC), a backdoor unlearning framework that eliminates the need for trigger recovery or poisoned sample identification. DBC operates in two phases. It first leverages catastrophic forgetting to eliminate the backdoor behavior. It then applies Sharpness-Aware Minimization (SAM) to flatten the loss landscape, improving generalization and restoring legitimate functionality without requiring prior knowledge of triggers or poisoned samples. To further enhance the flatness of the loss landscape, DBC employs an importance sampling strategy that prioritizes samples most sensitive to weight perturbations, effectively reducing the impact of poisoned instances. Extensive evaluations demonstrate that DBC neutralizes diverse backdoor attacks while maintaining model utility. Its superior performance achieved first place in the mitigation round of the TrojAI competition, a multi-year, multi-round benchmark for backdoor defense.

Data valuation provides a principled framework for quantifying the contribution of data to model training. It plays a crucial role in trustworthy machine learning (ML) by supporting data curation, enhancing interpretability, and enabling fair incentive mechanisms in data markets. Shapley value is a popular method for data valuation, but accurate estimation remains computationally expensive, particularly at the dataset level. In this paper, we introduce Ensemble Shapley, an efficient framework tailored for dataset-level valuation on the sharded structure. To reduce the computational costs, we propose a two-phase estimation method that apportions the intensive contribution computation costs across disjoint data shards and strategically reuses the computation results, achieving efficient contribution evaluation through the ensemble of shard models. However, weak shard models trained on noisy data may degrade ensemble models’ performance. To solve this, we introduce a behavior-driven guided sampling method that pairs noisy datasets with benign ones, ensuring reliable contribution estimates despite the noise. We also derive an advantageous lower bound for the number of evaluation iterations that balances efficiency and accuracy by the number of shards. Experimental results show Ensemble Shapley has superior efficiency over existing methods while maintaining comparable accuracy across various ML tasks, and demonstrates strong scalability and integration potential.