🏢 Warsaw University of Technology
Task-recency bias strikes back: Adapting covariances in Exemplar-Free Class Incremental Learning
·2323 words·11 mins·
loading
·
loading
Machine Learning
Continual Learning
🏢 Warsaw University of Technology
AdaGauss tackles task-recency bias in exemplar-free class incremental learning by adapting class covariances and introducing an anti-collapse loss, achieving state-of-the-art results.
Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion
·2629 words·13 mins·
loading
·
loading
Natural Language Processing
Large Language Models
🏢 Warsaw University of Technology
D2DMoE boosts Transformer efficiency by up to 60% via smart activation sparsity and dynamic expert selection, outperforming existing methods.
Bigger, Regularized, Optimistic: scaling for compute and sample efficient continuous control
·3405 words·16 mins·
loading
·
loading
Reinforcement Learning
🏢 Warsaw University of Technology
BRO (Bigger, Regularized, Optimistic) achieves state-of-the-art sample efficiency in continuous control by scaling critic networks and using strong regularization with optimistic exploration.