This paper explores the use of model-based offline reinforcement learning with long model rollouts. While some literature criticizes this approach due to compounding errors, many practitioners have found success in real-world applications. The paper aims to demonstrate that long rollouts do not necessarily result in exponentially growing errors and can actually produce better Q-value estimates than model-free methods. These findings can potentially enhance reinforcement learning techniques.
Learning with a Single Rollout via Pass@k Policy Optimization
University of Alberta（阿尔伯塔大学）；BIGAI（北京通用人工智能研究院）；University of Toronto（多伦多大学）；McGill University, Mila（麦吉尔大学，米拉）；Nvidia Research（英伟达研究院）；Amii（阿尔伯塔机器智能研究所）；CIFAR AI Chair（CIFAR人工智能教席）
Lookahead, also known as non-myopic, Bayesian optimization (BO) aims to find optimal sampling policies through solving a dynamic program (DP) that maximizes a long-term reward over a rolling horizon. Though promising, lookahead BO faces the risk of error propagation through its increased dependence on a possibly mis-specified model. In this work we focus on the rollout approximation for solving the intractable DP. We first prove the improving nature of rollout in tackling lookahead BO and provide a sufficient condition for the used heuristic to be rollout improving. We then provide both a theo…
University of Maryland（马里兰大学）；University of Colorado（科罗拉多大学）；Google Research Zürich（谷歌苏黎世研究中心）
Cross-lingual word embeddings (CLWE) are often evaluated on bilingual lexicon induction (BLI). Recent CLWE methods use linear projections, which underfit the training dictionary, to generalize on BLI. However, underfitting can hinder generalization to other downstream tasks that rely on words from the training dictionary. We address this limitation by retrofitting CLWE to the training dictionary, which pulls training translation pairs closer in the embedding space and overfits the training dictionary. This simple post-processing step often improves accuracy on two downstream tasks, despite low…
University of Oxford（牛津大学）；DeepMind（深度思考）；Facebook AI Research（Facebook人工智能研究院）
When observing the actions of others, humans make inferences about why they acted as they did, and what this implies about the world; humans also use the fact that their actions will be interpreted in this manner, allowing them to act informatively and thereby communicate efficiently with others. Although learning algorithms have recently achieved superhuman performance in a number of two-player, zero-sum games, scalable multi-agent reinforcement learning algorithms that can discover effective strategies and conventions in complex, partially observable settings have proven elusive. We present …
Gradient descent or its variants are popular in training neural networks. However, in deep Q-learning with neural network approximation, a type of reinforcement learning, gradient descent (also known as Residual Gradient (RG)) is barely used to solve Bellman residual minimization problem. On the contrary, Temporal Difference (TD), an incomplete gradient descent method prevails. In this work, we perform extensive experiments to show that TD outperforms RG, that is, when the training leads to a small Bellman residual error, the solution found by TD has a better policy and is more robust against …
FADE: Why Bad Descriptions Happen to Good Features
Department of Artificial Intelligence, Fraunhofer Heinrich Hertz Institute（人工智能系，弗劳恩霍夫海因里希·赫兹研究所）；Department of Electrical Engineering and Computer Science, Technische Universität Berlin（电气工程与计算机科学系，柏林技术大学）；BIFOLD - Berlin Institute for the Foundations of Learning and Data（柏林学习与数据基础研究所）；Centre of eXplainable Artificial Intelligence, Technological University Dublin（可解释人工智能中心，都柏林技术大学）
Columbia University（哥伦比亚大学）；Stanford University（斯坦福大学）
We use smoothed analysis techniques to provide guarantees on the training loss of Multilayer Neural Networks (MNNs) at differentiable local minima. Specifically, we examine MNNs with piecewise linear activation functions, quadratic loss and a single output, under mild over-parametrization. We prove that for a MNN with one hidden layer, the training error is zero at every differentiable local minimum, for almost every dataset and dropout-like noise realization. We then extend these results to the case of more than one hidden layer. Our theoretical guarantees assume essentially nothing on the tr…
Model-based planning is often thought to be necessary for deep, careful reasoning and generalization in artificial agents. While recent successes of model-based reinforcement learning (MBRL) with deep function approximation have strengthened this hypothesis, the resulting diversity of model-based methods has also made it difficult to track which components drive success and why. In this paper, we seek to disentangle the contributions of recent methods by focusing on three questions: (1) How does planning benefit MBRL agents? (2) Within planning, what choices drive performance? (3) To what exte…
Institut de physique théorique, Université Paris Saclay, CNRS, CEA（巴黎萨克雷大学理论物理研究所，法国国家科学研究中心，法国原子能和替代能源委员会）；Laboratoire de Physique de l’Ecole normale supérieure ENS, Université PSL, CNRS, Sorbonne Université, Université Paris-Diderot（巴黎高等师范学院物理实验室，巴黎文理研究大学，法国国家科学研究中心，索邦大学，巴黎狄德罗大学）；King’s College London（伦敦国王学院）
Gradient-based algorithms are effective for many machine learning tasks, but despite ample recent effort and some progress, it often remains unclear why they work in practice in optimising high-dimensional non-convex functions and why they find good minima instead of being trapped in spurious ones. Here we present a quantitative theory explaining this behaviour in a spiked matrix-tensor model. Our framework is based on the Kac-Rice analysis of stationary points and a closed-form analysis of gradient-flow originating from statistical physics. We show that there is a well defined region of param…
Within the last few years, there has been a move towards using statistical models in conjunction with neural networks with the end goal of being able to better answer the question, "what do our models know?". From this trend, classical metrics such as Prediction Interval Coverage Probability (PICP) and new metrics such as calibration error have entered the general repertoire of model evaluation in order to gain better insight into how the uncertainty of our model compares to reality. One important component of uncertainty modeling is model uncertainty (epistemic uncertainty), a measurement of …
