Coresets are More Than Replay: A Data-Centric View of Continual Learning
Shows that coreset selection can be applied to the whole training set, not just a small replay buffer, substantially cutting the amount of data needed for continual learning.
All Publications
Pruned Adaptation Modules (PAM): A Simple yet Strong Baseline for Continual Foundation Models
Keeps a pre-trained backbone frozen and adds small sparse task-specific adapters, giving a lightweight yet strong baseline for continual learning with foundation models.
Unlocking [CLS] Features for Continual Post-Training (TOSCA)
Introduces TOSCA, a sparse adapter-calibrator module on the classification token that lets foundation models keep learning new tasks without losing prior knowledge.
Continual Learning with Dynamic Sparse Training: Exploring Algorithms for Effective Model Updates
Explores dynamic sparse training algorithms as an efficient way to update models continually without full retraining.
AdaCL: Adaptive Continual Learning
Uses Bayesian optimization to automatically tune each new task's hyperparameters, improving accuracy and reducing forgetting in class-incremental learning.
Automated Machine Learning Approach in Material Discovery of Hole Selective Layers for Perovskite Solar Cells
Applies automated machine learning to discover new hole-selective layer materials for perovskite solar cells.
Predicting Perovskite Bandgap and Solar Cell Performance with Machine Learning
Uses machine learning models to predict perovskite bandgap and solar cell performance from materials data.
A Machine Learning Approach for Metal Oxide Based Polymer Composites as Charge Selective Layers in Perovskite Solar Cells
Applies machine learning to evaluate metal-oxide polymer composites as charge-selective layers in perovskite solar cells.
SMOTE-NC and Gradient Boosting Imputation Based Random Forest Classifier for Predicting Severity Level of COVID-19 Patients with Blood Samples
Combines SMOTE-NC oversampling and gradient boosting imputation with a random forest classifier to predict COVID-19 severity from blood sample data.
Comparison of Machine Learning Models on Performance of Single- and Dual-Type Electrochromic Devices
Compares several machine learning models for predicting the performance of single- and dual-type electrochromic devices.