用测试时训练校准大模型推理置信度，让 conformal 方法在分布变化下依然可靠，COLM 2026 新思路值得关注。
arXiv:2604.01170v2 Announce Type: replace-cross Abstract: While test-time scaling has enabled large language models to solve highly difficult tasks, s…
提出“思维策略”框架，通过在线策略进化实现测试时训练，打破冻结策略限制，显著提升大模型复杂推理能力。
arXiv:2601.20379v2 Announce Type: replace Abstract: Large language models (LLMs) struggle with complex, long-horizon reasoning due to instability caus…
自我引导的测试时训练，专治长上下文LLM的准确率退化难题。
arXiv:2607.09415v1 Announce Type: cross Abstract: Long-context processing has become increasingly important for large language models (LLMs), but simp…
提出Agentic Test-Time Training方法，让LLM智能体在长时任务中自适应调整权重，有效解决轨迹退化、策略失效难题。
arXiv:2607.03441v1 Announce Type: cross Abstract: LLM agents often degrade over long episodes: as trajectories grow, they revisit explored states, rep…
测试时训练宣称能记住部署数据，但困惑度真的够吗？本文提出行为评估框架，用交互式测试验证模型记忆声明，比指标更可信。
arXiv:2607.00368v1 Announce Type: new Abstract: Large language model test-time training (TTT) is often evaluated through local proxy metrics: models a…
挑战测试时训练新范式：用语言模型自监督的下一词预测信号作为内循环目标，无需额外损失函数。
arXiv:2606.21803v1 Announce Type: new Abstract: Next-token prediction is the self-supervised signal that trains language models, and every observed pr…
探索测试时训练在近似采样中的强大作用，为生成式AI推理难题提供新思路。
arXiv:2606.11437v1 Announce Type: cross Abstract: Efficiently sampling from a complex probability distribution is a fundamental problem which has beco…
提出测试时训练新范式，将Vision Transformer复杂度从二次降至线性，推理时仅需少量计算即可大幅提升准确率和鲁棒性。
arXiv:2605.02772v2 Announce Type: replace Abstract: While linear-complexity attention mechanisms offer a promising alternative to Softmax attention fo…
测试时训练首次应用于监督因果学习，破解分布外泛化难题的新范式。
arXiv:2605.30015v1 Announce Type: cross Abstract: Supervised Causal Learning (SCL) has shown promise in causal discovery by framing it as a supervised…
最新研究揭示：测试时训练（TTT）能绕过AI安全护栏，引发对模型防御机制的新思考。
arXiv:2605.22984v1 Announce Type: cross Abstract: Test-Time Training (TTT) is an emerging paradigm that enables models to adapt their parameters durin…