test-time-adaptation

This paper presents a theoretical framework for self-poisoning in adaptive out-of-distribution detection, proving a sharp threshold for collapse and a certified label-free calibration method that severs the feedback loop. It also provides impossibility results for distinguishing drift from contamination without labels.

OPERA proposes a multi-agent ensemble framework that treats expert weight assignment as an offline policy learning problem for universal biomedical image analysis, enabling test-time adaptation without retraining and consistently improving performance across 9 datasets and 30+ baselines.

Black-Mamba introduces a test-time adaptive forecasting architecture that uses accumulated surprisal to selectively update memory only upon evidence of distribution drift, achieving efficient adaptation on non-stationary time series.

AdaJEPA introduces an adaptive latent world model that continuously updates during test-time via closed-loop model predictive control, significantly improving planning success under distribution shift.

Introduces Reward-Gated Test-Time Adaptation (RG-TTA), a reinforcement learning framework that selectively applies debiasing to CLIP models based on input bias sensitivity, resolving the fairness-utility trade-off.

Proposes BP-TTA, a test-time adaptation method that handles both class imbalance and continual domain shifts by combining batch-balanced sampling with prototype-guided constraints, achieving state-of-the-art performance in dynamic streaming scenarios.

ComMem proposes complementary memory systems inspired by biological memory to improve test-time adaptation of vision-language models, outperforming state-of-the-art on 15 benchmarks.

This paper proposes a test-time adaptation approach using semi-supervised learning for AI text detection that adapts to continual distribution shifts from new LLMs, adversarial humanization, and temporal drift, outperforming state-of-the-art supervised detectors.

QGF is an RL algorithm that improves policies at test time by using a value gradient to guide a pre-trained flow policy, avoiding training-time instability while maintaining competitive performance.

Proposes Demo2Reward, a test-time prompt optimization technique for VLM reward models using a few expert demonstrations, significantly reducing false positives and improving policy learning in robotics without additional model training.

This paper develops a PAC-Bayesian framework for test-time adaptation that uses MMD-balls as credal sets, providing formal generalization bounds and separating epistemic from aleatoric uncertainty under distribution shift.

Proposes a hierarchical variational policy framework for reward-guided diffusion, enabling high-quality sampling with reduced inference cost. Achieves strong quality-speed tradeoff on tasks like super-resolution.

SOLAR proposes a self-optimizing autonomous agent that leverages parameter-level meta-learning and multi-level reinforcement learning to enable lifelong adaptation of LLMs to non-stationary data streams, outperforming baselines on reasoning tasks.

Proposes Federated Nested Learning (FedNL), a framework that reformulates federated learning as a three-level nested optimization system, enabling collaborative training of self-referential memories for test-time adaptation to handle Non-IID data and long-tail distributions.

This paper proposes RMemSafe, a reliability-gated extension for continual test-time adaptation that attenuates source anchoring when the frozen source's predictive entropy becomes high, preventing blind anchoring under source collapse. The method achieves state-of-the-art error reduction on the CCC benchmark.

This paper introduces TacoMAS, a framework for test-time co-evolution of agent capabilities and communication topology in LLM-based multi-agent systems. It demonstrates that jointly adapting fast capability loops and slow topology loops improves performance and stability over existing baselines.

FAAST proposes a forward-only method that compiles labeled examples into fast weights analytically, enabling efficient test-time supervised adaptation without backpropagation, achieving over 90% speedup and 95% memory savings while maintaining performance.

This paper proposes CAP-TTA, a test-time adaptation framework that uses preconditioned LoRA updates triggered by bias-risk scores to mitigate toxicity and bias in large language models during narrative generation, achieving faster optimization and better fluency than standard baselines.

TTL introduces a test-time textual learning framework for OOD detection using pretrained vision-language models like CLIP, which dynamically learns OOD semantics from unlabeled test streams without external OOD labels. The method uses pseudo-labeled samples and an OOD knowledge purification strategy to improve detection robustness across diverse and evolving OOD distributions.