Tag
论文提出 LSA-PAST，一种面向组合式增量学习（composition-IL）的参数高效方法，通过在冻结 ViT 中注入深度特化的 LoRA 适配器并用无复习的仿射变换将旧原型映射到新适配器空间，实现了精度-遗忘权衡的 state-of-the-art 表现。
This paper introduces StreamFraudNet, a weakly supervised incremental model for detecting phone scams from raw speech, achieving a ROC-AUC of 0.9953 and operating in real-time to provide early warnings during calls.
The paper studies training dynamics for two-layer neural networks learning orthogonal multi-index models under standard initialization, revealing incremental learning and competitive parameter reallocation, and introduces a symmetry-based finite-width approximation for analysis.
This paper introduces a provenance-guided incremental learning framework to handle rule-induced concept shifts, where target definitions change, and evaluates it on a new benchmark with improved efficiency and accuracy.
SynGAP is a task-free continual learning framework that simulates biological metaplasticity via adaptive gradient preconditioning to mitigate catastrophic forgetting, demonstrating significant accuracy improvements over existing methods on benchmarks.
This paper introduces a quantum incremental learning framework using trainable mixed-state prototypes, enabling new classes to be added without increasing circuit width while mitigating catastrophic forgetting.
DeltaMem organizes LLM agent memory into residual trees to reduce redundancy and retrieval conflicts, storing incremental variations of experiences for continual learning.
This paper investigates adversarial robustness in Fuzzy ARTMAP, a streaming neural architecture, by introducing WB-Softmax as a mechanism-aligned white-box attack surrogate. It evaluates progressive training and selective updating strategies to improve robustness without data replay, while also offering interpretable diagnostics for structural failures.