As previous representations for reinforcement learning cannot effectively incorporate a human-intuitive understanding of the 3D environment, they usually suffer from sub-optimal performances. In this paper, we present Semantic-aware Neural Radiance Fields for Reinforcement Learning (SNeRL), which jointly optimizes semantic-aware neural radiance fields (NeRF) with a convolutional encoder to learn 3D-aware neural implicit representation from multi-view images. We introduce 3D semantic and distilled feature fields in parallel to the RGB radiance fields in NeRF to learn semantic and object-centric representation for reinforcement learning. SNeRL outperforms not only previous pixel-based representations but also recent 3D-aware representations both in model-free and model-based reinforcement learning.

Welcome to our May 2023 monthly digest, where you can catch up with any AIhub stories you may have missed, peruse the latest news, find out about recent events, and more. This month, we learn how to mitigate biases in machine learning, explore tradeoffs in school redistricting, and find out how machine learning algorithms fared in predicting the winner of this year's Eurovision Song Contest. In this blogpost, Max Springer examines the notion of fairness in hierarchical clustering. Max and colleagues demonstrate that it's possible to incorporate fairness constraints or demographic information into the optimization process to reduce biases in ML models without significantly sacrificing performance. Joar Skalse and Alessandro Abate won the AAAI 2023 outstanding paper award for their work, Misspecification in Inverse Reinforcement Learning, in which they study the question of how robust the inverse reinforcement learning problem is to misspecification of the underlying behavioural model.

The rapid advancement of artificial intelligence (AI) systems suggests that artificial general intelligence (AGI) systems may soon arrive. Many researchers are concerned that AIs and AGIs will harm humans via intentional misuse (AI-misuse) or through accidents (AI-accidents). In respect of AI-accidents, there is an increasing effort focused on developing algorithms and paradigms that ensure AI systems are aligned to what humans intend, e.g. AI systems that yield actions or recommendations that humans might judge as consistent with their intentions and goals. Here we argue that alignment to human intent is insufficient for safe AI systems and that preservation of long-term agency of humans may be a more robust standard, and one that needs to be separated explicitly and a priori during optimization. We argue that AI systems can reshape human intention and discuss the lack of biological and psychological mechanisms that protect humans from loss of agency. We provide the first formal definition of agency-preserving AI-human interactions which focuses on forward-looking agency evaluations and argue that AI systems - not humans - must be increasingly tasked with making these evaluations. We show how agency loss can occur in simple environments containing embedded agents that use temporal-difference learning to make action recommendations. Finally, we propose a new area of research called "agency foundations" and pose four initial topics designed to improve our understanding of agency in AI-human interactions: benevolent game theory, algorithmic foundations of human rights, mechanistic interpretability of agency representation in neural-networks and reinforcement learning from internal states.

The optimal multicast tree problem in the Software-Defined Networking (SDN) multicast routing is an NP-hard combinatorial optimization problem. Although existing SDN intelligent solution methods, which are based on deep reinforcement learning, can dynamically adapt to complex network link state changes, these methods are plagued by problems such as redundant branches, large action space, and slow agent convergence. In this paper, an SDN intelligent multicast routing algorithm based on deep hierarchical reinforcement learning is proposed to circumvent the aforementioned problems. First, the multicast tree construction problem is decomposed into two sub-problems: the fork node selection problem and the construction of the optimal path from the fork node to the destination node. Second, based on the information characteristics of SDN global network perception, the multicast tree state matrix, link bandwidth matrix, link delay matrix, link packet loss rate matrix, and sub-goal matrix are designed as the state space of intrinsic and meta controllers. Then, in order to mitigate the excessive action space, our approach constructs different action spaces at the upper and lower levels. The meta-controller generates an action space using network nodes to select the fork node, and the intrinsic controller uses the adjacent edges of the current node as its action space, thus implementing four different action selection strategies in the construction of the multicast tree. To facilitate the intelligent agent in constructing the optimal multicast tree with greater speed, we developed alternative reward strategies that distinguish between single-step node actions and multi-step actions towards multiple destination nodes.

Via operator theoretic methods, we formalize the concentration phenomenon for a given observable `$r$' of a discrete time Markov chain with `$\mu_{\pi}$' as invariant ergodic measure, possibly having support on an unbounded state space. The main contribution of this paper is circumventing tedious probabilistic methods with a study of a composition of the Markov transition operator $P$ followed by a multiplication operator defined by $e^{r}$. It turns out that even if the observable/ reward function is unbounded, but for some for some $q>2$, $\|e^{r}\|_{q \rightarrow 2} \propto \exp\big(\mu_{\pi}(r) +\frac{2q}{q-2}\big) $ and $P$ is hyperbounded with norm control $\|P\|_{2 \rightarrow q }< e^{\frac{1}{2}[\frac{1}{2}-\frac{1}{q}]}$, sharp non-asymptotic concentration bounds follow. \emph{Transport-entropy} inequality ensures the aforementioned upper bound on multiplication operator for all $q>2$. The role of \emph{reversibility} in concentration phenomenon is demystified. These results are particularly useful for the reinforcement learning and controls communities as they allow for concentration inequalities w.r.t standard unbounded obersvables/reward functions where exact knowledge of the system is not available, let alone the reversibility of stationary measure.

This paper is concerned with the problem of policy evaluation with linear function approximation in discounted infinite horizon Markov decision processes. We investigate the sample complexities required to guarantee a predefined estimation error of the best linear coefficients for two widely-used policy evaluation algorithms: the temporal difference (TD) learning algorithm and the two-timescale linear TD with gradient correction (TDC) algorithm. In both the on-policy setting, where observations are generated from the target policy, and the off-policy setting, where samples are drawn from a behavior policy potentially different from the target policy, we establish the first sample complexity bound with high-probability convergence guarantee that attains the optimal dependence on the tolerance level. We also exhihit an explicit dependence on problem-related quantities, and show in the on-policy setting that our upper bound matches the minimax lower bound on crucial problem parameters, including the choice of the feature maps and the problem dimension.

A common paradigm for Guided sampling is a vital approach for applying introducing human preference in diffusion models is guided diffusion models in real-world tasks that embeds sampling, which includes classifier guidance (Dhariwal & human-defined guidance during the sampling procedure. Nichol, 2021), classifier-free guidance (Ho & Salimans, This paper considers a general setting 2021) and other guidance methods (Nichol et al., 2021; Ho where the guidance is defined by an (unnormalized) et al., 2022c; Zhao et al., 2022). By leveraging guided sampling, energy function. The main challenge for diffusion models can realize amazing text-to-image this setting is that the intermediate guidance during generation (Saharia et al., 2022b), video generation (Ho the diffusion sampling procedure, which is et al., 2022c;a; Yang et al., 2022; Zhou et al., 2022), controllable jointly defined by the sampling distribution and text generation (Li et al., 2022), inverse molecular the energy function, is unknown and is hard to design (Bao et al., 2022b) and reinforcement learning (Janner estimate. To address this challenge, we propose et al., 2022; Chen et al., 2022; Ajay et al., 2022).

Despite the recent success of representation learning in sequential decision making, the study of the pure exploration scenario (i.e., identify the best option and minimize the sample complexity) is still limited. In this paper, we study multi-task representation learning for best arm identification in linear bandits (RepBAI-LB) and best policy identification in contextual linear bandits (RepBPI-CLB), two popular pure exploration settings with wide applications, e.g., clinical trials and web content optimization. In these two problems, all tasks share a common low-dimensional linear representation, and our goal is to leverage this feature to accelerate the best arm (policy) identification process for all tasks. For these problems, we design computationally and sample efficient algorithms DouExpDes and C-DouExpDes, which perform double experimental designs to plan optimal sample allocations for learning the global representation. We show that by learning the common representation among tasks, our sample complexity is significantly better than that of the native approach which solves tasks independently. To the best of our knowledge, this is the first work to demonstrate the benefits of representation learning for multi-task pure exploration.

Specifying reward signals that allow agents to learn complex behaviors is a long-standing challenge in reinforcement learning. A promising approach is to extract preferences for behaviors from unlabeled videos, which are widely available on the internet. We present Video Prediction Rewards (VIPER), an algorithm that leverages pretrained video prediction models as action-free reward signals for reinforcement learning. Specifically, we first train an autoregressive transformer on expert videos and then use the video prediction likelihoods as reward signals for a reinforcement learning agent. VIPER enables expert-level control without programmatic task rewards across a wide range of DMC, Atari, and RLBench tasks. Moreover, generalization of the video prediction model allows us to derive rewards for an out-of-distribution environment where no expert data is available, enabling cross-embodiment generalization for tabletop manipulation. We see our work as starting point for scalable reward specification from unlabeled videos that will benefit from the rapid advances in generative modeling. Source code and datasets are available on the project website: https://escontrela.me/viper

Actor-critic algorithms address the dual goals of reinforcement learning (RL), policy evaluation and improvement, via two separate function approximators. The practicality of this approach comes at the expense of training instability, caused mainly by the destructive effect of the approximation errors of the critic on the actor. We tackle this bottleneck by employing an existing Probably Approximately Correct (PAC) Bayesian bound for the first time as the critic training objective of the Soft Actor-Critic (SAC) algorithm. We further demonstrate that online learning performance improves significantly when a stochastic actor explores multiple futures by critic-guided random search. We observe our resulting algorithm to compare favorably to the state of the art on multiple classical control and locomotion tasks in terms of both sample efficiency and regret minimization.