^abMeta-learners for Estimating Heterogeneous Treatment Effects using Machine Learning https://arxiv.org/abs/1706.03461 ^abQuasi-Oracle Estimation of Heterogeneous Treatment Effects https://arxiv.org/abs/1712.04912 ^Root-N-Consistent Semiparametric Regression https://www.jstor.org/stable/1912705 ...
4. meta-learners优化:当得到线上推理数据的完整标签后,使用Ltest对DA和MA进行更新,公式如下 ϕk,ψk=argminϕk,ψkLtest(D~testk,θk),这里的θk=MA(D~traink;θk−1)
在强化学习领域，meta-learning同样发挥着重要作用。在需要学习如何与环境互动以最大化奖励的任务中，meta-learners能够快速适应新环境，并通过有限的探索学习到最优策略。这一特性使得meta-learning在解决复杂和多样的强化学习问题中具有独特的优势。然而，尽管meta-learning具有巨大的潜力，但它仍面临着一些挑战。首先，样...
However, the tasks used to evaluate these meta-learners frequently diverge from the problems in the real world that they are meant to resolve. This work aims to apply meta-learning to a problem that is more pertinent to the real world: class incremental learning (IL). In this scenario, ...
[117]. When computing task similarity, it is also important to normalize all meta-features [9], perform feature selection [172], or employ dimensionality reduction techniques (e.g. PCA) [17]. When learning meta-models, one can also use relational meta-learners [173] or case-based ...
Language Models are Unsupervised Multitask Learners (2019)过去，训练 LLM 涉及监督学习，需要大量针对所需任务的标记数据。本文探讨了无监督学习的潜力，其中模型从大量未标记的文本数据中学习。通过对大量未标记的文本数据进行训练，LLM 可以自然而然地学会执行各种任务（多任务学习），而无需明确的任务特定监督。这种...
2021 年，Girshick 参与的论文《Masked Autoencoders Are Scalable Vision Learners》成为了计算机视觉圈的热门话题。这篇论文展示了一种被称为掩码自编码器（masked autoencoders，MAE）的新方法，可以用作计算机视觉的可扩展自监督学习器。今年，Meta 发布了「分割一切」（Segment Anything）模型（SAM），被很多人誉为...
(2019). Metalearners for estimating heterogeneous treatment effects using machine learning. Proceedings of the National Academy of Sciences, 116(10), 4156–4165. 模型 T-Learner T是two的缩写,是比较传统的ML模型用于因果推理的方式。对照组和实验组进行分别建模得到两个模型,对每个样本计算两个模型的预测值...
Few-shot Learning with Meta Metric Learners Yu Cheng, Mo Yu, Xiaoxiao Guo, Bowen Zhou --NIPS 2017 workshop on Meta-Learning Learning Embedding Adaptation for Few-Shot Learning [paper] Han-Jia Ye, Hexiang Hu, De-Chuan Zhan, Fei Sha --arXiv 2018 ...
Meta-learners available: warp(requires thesrc/warpgradpackage) leap(requires thesrc/leappackage) reptile fomaml(first-order MAML) maml(requires thesrc/mamlpackage) ft(multi-headed finetuning) no(no meta-training) Dev. tips Default configs are for optimal performance. If want to take the code ...