Dyna-style reinforcement learning (RL) agents improve sample efficiency over model-free RL agents by updating the value function with simulated experience generated by an environment model. However, it is often difficult to learn accurate models of environment dynamics, and even small errors may result in failure of Dyna agents. In this paper, we highlight that one potential cause of that failure is bootstrapping off of the values of simulated states, and introduce a new Dyna algorithm to avoid this failure. We discuss a design space of Dyna algorithms, based on using successor or predecessor models---simulating forwards or backwards---and using one-step or multi-step updates. Three of the variants have been explored, but surprisingly the fourth variant has not: using predecessor models with multi-step updates. We present the \emph{Hallucinated Value Hypothesis} (HVH): updating the values of real states towards values of simulated states can result in misleading action values which adversely affect the control policy. We discuss and evaluate all four variants of Dyna amongst which three update real states toward simulated states --- so potentially toward hallucinated values --- and our proposed approach, which does not. The experimental results provide evidence for the HVH, and suggest that using predecessor models with multi-step updates is a fruitful direction toward developing Dyna algorithms that are more robust to model error.

Recent work has highlighted the label alignment property (LAP) in supervised learning, where the vector of all labels in the dataset is mostly in the span of the top few singular vectors of the data matrix. Drawing inspiration from this observation, we propose a regularization method for unsupervised domain adaptation that encourages alignment between the predictions in the target domain and its top singular vectors. Unlike conventional domain adaptation approaches that focus on regularizing representations, we instead regularize the classifier to align with the unsupervised target data, guided by the LAP in both the source and target domains. Theoretical analysis demonstrates that, under certain assumptions, our solution resides within the span of the top right singular vectors of the target domain data and aligns with the optimal solution. By removing the reliance on the commonly used optimal joint risk assumption found in classic domain adaptation theory, we showcase the effectiveness of our method on addressing problems where traditional domain adaptation methods often fall short due to high joint error. Additionally, we report improved performance over domain adaptation baselines in well-known tasks such as MNIST-USPS domain adaptation and cross-lingual sentiment analysis. An implementation is available at https://github.com/EhsanEI/lar/.

Deep neural networks are widely known for their remarkable effectiveness across various tasks, with the consensus that deeper networks implicitly learn more complex data representations. This paper shows that sufficiently deep networks trained for supervised image classification split into two distinct parts that contribute to the resulting data representations differently. The initial layers create linearly-separable representations, while the subsequent layers, which we refer to as \textit{the tunnel}, compress these representations and have a minimal impact on the overall performance. We explore the tunnel's behavior through comprehensive empirical studies, highlighting that it emerges early in the training process. Its depth depends on the relation between the network's capacity and task complexity. Furthermore, we show that the tunnel degrades out-of-distribution generalization and discuss its implications for continual learning.

A variety of theoretically-sound policy gradient algorithms exist for the on-policy setting due to the policy gradient theorem, which provides a simplified form for the gradient. The off-policy setting, however, has been less clear due to the existence of multiple objectives and the lack of an explicit off-policy policy gradient theorem. In this work, we unify these objectives into one off-policy objective, and provide a policy gradient theorem for this unified objective. The derivation involves emphatic weightings and interest functions. We show multiple strategies to approximate the gradients, in an algorithm called Actor Critic with Emphatic weightings (ACE). We prove in a counterexample that previous (semi-gradient) off-policy actor-critic methods—particularly Off-Policy Actor-Critic (OffPAC) and Deterministic Policy Gradient (DPG)—converge to the wrong solution whereas ACE finds the optimal solution. We also highlight why these semi-gradient approaches can still perform well in practice, suggesting strategies for variance reduction in ACE. We empirically study several variants of ACE on two classic control environments and an image-based environment designed to illustrate the tradeoffs made by each gradient approximation. We find that by approximating the emphatic weightings directly, ACE performs as well as or better than OffPAC in all settings tested.

For multi-valued functions---such as when the conditional distribution on targets given the inputs is multi-modal---standard regression approaches are not always desirable because they provide the conditional mean. Modal regression algorithms address this issue by instead finding the conditional mode(s). Most, however, are nonparametric approaches and so can be difficult to scale. Further, parametric approximators, like neural networks, facilitate learning complex relationships between inputs and targets. In this work, we propose a parametric modal regression algorithm. We use the implicit function theorem to develop an objective, for learning a joint function over inputs and targets. We empirically demonstrate on several synthetic problems that our method (i) can learn multi-valued functions and produce the conditional modes, (ii) scales well to high-dimensional inputs, and (iii) can even be more effective for certain uni-modal problems, particularly for high-frequency functions. We demonstrate that our method is competitive in a real-world modal regression problem and two regular regression datasets.

There is growing evidence that converting targets to soft targets in supervised learning can provide considerable gains in performance. Much of this work has considered classification, converting hard zero-one values to soft labels—such as by adding label noise, incorporating label ambiguity or using distillation. In parallel, there is some evidence from a regression setting in reinforcement learning that learning distributions can improve performance. In this work, we investigate the reasons for this improvement, in a regression setting. We introduce a novel distributional regression loss, and similarly find it significantly improves prediction accuracy. We investigate several common hypotheses, around reducing overfitting and improved representations. We instead find evidence for an alternative hypothesis: this loss is easier to optimize, with better behaved gradients, resulting in improved generalization. We provide theoretical support for this alternative hypothesis, by characterizing the norm of the gradients of this loss.

Policy gradient methods are widely used for control in reinforcement learning, particularly for the continuous action setting. There have been a host of theoretically sound algorithms proposed for the on-policy setting, due to the existence of the policy gradient theorem which provides a simplified form for the gradient. In off-policy learning, however, where the behaviour policy is not necessarily attempting to learn and follow the optimal policy for the given task, the existence of such a theorem has been elusive. In this work, we solve this open problem by providing the first off-policy policy gradient theorem. The key to the derivation is the use of emphatic weightings. We develop a new actor-critic algorithm—called Actor Critic with Emphatic weightings (ACE)—that approximates the simplified gradients provided by the theorem. We demonstrate in a simple counterexample that previous off-policy policy gradient methods—particularly OffPAC and DPG—converge to the wrong solution whereas ACE finds the optimal solution.

In many real-world applications, data contain heterogeneous input modalities (e.g., web pages include images, text, etc.). Moreover, data such as images are usually described using different views (i.e. different sets of features). Learning a distance metric or similarity measure that originates fr om all input modalities or views is essential for many tasks such as content-based retrieval ones. In these cases, similar and dissimilar pairs of data can be used to find a better representation of data in which similarity and dissimilarity constraints are better satisfied. In this paper, we incorporate supervision in the form of pairwise similarity and/or dissimilarity constraints into multi-modal deep networks to combine different modalities into a shared latent space. Using properties of multi-modal data, we design multi-modal deep networks and propose a pre-training algorithm for these networks. In fact, the proposed network has the ability of learning intra- and inter-modal high-order statistics from raw features and we control its high flexibility via an efficient multi-stage pre-training phase corresponding to properties of multi-modal data. Experimental results show that the proposed method outperforms recent methods on image retrieval tasks.