Entity · technique

stability-plasticity dilemma

techniqueactivestability-plasticity-dilemma-c93f9c7e·1 events·first seen May 28, 2026

Aliases: stability-plasticity dilemma

Merged from

Co-occurring entities

orthogonal finetuning PEFT-Arena Parameter-Efficient Fine-Tuning path-wise rewinding supervised fine-tuning

More like this (12)

Inference-time Plasticity Preserving Plasticity in Continual Learning via Dynamical Isometry Dynamic-Probabilistic Consistency Gap The Stable Recovery Manifold: Geometric Principles Governing Recoverability in Continual Learning adaptive thinking Pessimism's Paradox: Conservative Offline Training Amplifies Reward Hacking During Online Adaptation in Reasoning Models epistemic uncertainty Moravec's Paradox stable-worldmodel Structural Tension From Application-Layer Simulation to Native Meta-Architecture: Structural Tension as an Endogenous Driver for Heterogeneous AI Evolution Consistency Training Can Entrench Misalignment

Recent events (1)

6arXiv · cs.LG·May 28, 2026·source ↗

PEFT-Arena: Benchmarking Parameter-Efficient Finetuning via Stability-Plasticity Trade-offs

PEFT-Arena is a new benchmark that evaluates parameter-efficient finetuning methods jointly on downstream task performance and retention of pretrained general capabilities, framing the problem as a stability-plasticity dilemma. Across methods tested under comparable parameter budgets, orthogonal finetuning achieves the best Pareto frontier. The paper provides geometric analyses in both weight space (spectral/singular-value structure) and activation space (representation distortion metrics) to explain why different PEFT methods differ in forgetting behavior. A practical finding is that final SFT checkpoints often overshoot an optimal retention operating point, motivating path-wise rewinding as a post-hoc correction.

Evaluation and Benchmarking Agent and Tool Ecosystem stability-plasticity dilemma orthogonal finetuning PEFT-Arena +4 more