The paper re-interprets a family of generative statistical models under the framework of sequential decision making used in reinforcement learning. It develops connections between training algorithms from the two fields. Models following this framework are tested on image datasets and compared to a baseline. The paper proposes a new and interesting view of generative models under the light of sequential decision making. This work clearly opens new perspectives and should allow the formulation of new ways to address learning and inference problems in a broad class of statistical models.

The exact mechanic of the policy transfer between different algorithm is not given. Given the content, I may assume that "transfer" means a simple copying of the parameters, but I remain unsure. When augmenting the experience buffer with other algorithm, it would be nice to clarify why it does (not) introduce any bias in the data. It seems that the different parts of the framework could be replaced by a different way of "tinkering" with a algorithm or its hyperparameters. E.g., the auxiliary on-policy algorithms are here mainly for exploration, but the exploration of the main off-policy algorithm itself can be easily controlled and I suspect it can, with the right setting, work as good as the given complicated framework. The global and local experience buffer seems more like a hack.

Following the rebuttals, all four reviewers agreed that this paper should be accepted. While there are remaining questions around the hyperparameters (and performance relative to other methods), and computational cost, this is an interesting and novel line of work. The authors are encouraged to proofread the paper thoroughly and address the issues raised by the reviewers.

In this work, we present BiM-ACPPO, a bilevel multi-armed bandit-based hierarchical reinforcement learning framework for interaction-aware decision-making and planning at unsignalized intersections. Essentially, it proactively takes the uncertainties associated with surrounding vehicles (SVs) into consideration, which encompass those stemming from the driver's intention, interactive behaviors, and the varying number of SVs. Intermediate decision variables are introduced to enable the high-level RL policy to provide an interaction-aware reference, for guiding low-level model predictive control (MPC) and further enhancing the generalization ability of the proposed framework. By leveraging the structured nature of self-driving at unsignalized intersections, the training problem of the RL policy is modeled as a bilevel curriculum learning task, which is addressed by the proposed Exp3.S-based BiMAB algorithm. It is noteworthy that the training curricula are dynamically adjusted, thereby facilitating the sample efficiency of the RL training process. Comparative experiments are conducted in the high-fidelity CARLA simulator, and the results indicate that our approach achieves superior performance compared to all baseline methods. Furthermore, experimental results in two new urban driving scenarios clearly demonstrate the commendable generalization performance of the proposed method.

Abstract--Advances in artificial intelligence (AI) including foundation models (FMs), are increasingly transforming human society, with smart city driving the evolution of urban living. Meanwhile, vehicle crowdsensing (VCS) has emerged as a key enabler, leveraging vehicles' mobility and sensor-equipped capabilities. In particular, ride-hailing vehicles can effectively facilitate flexible data collection and contribute towards urban intelligence, despite resource limitations. Therefore, this work explores a promising scenario, where edge-assisted vehicles perform joint tasks of order serving and the emerging foundation model finetuning using various urban data. However, integrating the VCS AI task with the conventional order serving task is challenging, due to their inconsistent spatio-temporal characteristics: (i) The distributions of ride orders and data point-of-interests (PoIs) may not coincide in geography, both following a priori unknown patterns; (ii) they have distinct forms of temporal effects, i.e., prolonged waiting makes orders become instantly invalid while data with increased staleness gradually reduces its utility for model fine-tuning. To overcome these obstacles, we propose an online framework based on multi-agent reinforcement learning (MARL) with careful augmentation. A new quality-of-service (QoS) metric is designed to characterize and balance the utility of the two joint tasks, under the effects of varying data volumes and staleness. Each RSU, equipped with a server, stores a complete base model, enabling vehicles to perform real-time fine-tuning as they collect data and transfer the I. X. Zhang are with the School of Computer Science and A previous version appears at IWQoS 2024 as a short paper. Due to the large volume, data stored in the government agencies in better city management. Notably, ridehailing RSU server can be discarded in a certain period of time. In vehicles are particularly advantageous for VCS tasks, practice, these data can be descriptive features and feedbacks due to their centralized ride-hailing platform management, (labels) of recommendation or generative AR applications, which reduces the cost of deploying and executing crowdsensing generated by nearby visitors or residents. They can also be tasks, and utilizes the data and computing resources traffic/environment monitoring data with labels generated by from ride-hailing vehicles to maximize the VCS task utilities. The government or any company that collaborates model (FM)-powered AI applications have revolutionized with the ride-hailing vehicle company has multiple types of numerous aspects of human lives, including healthcare, education, VSC tasks to fulfill, each of which needs certain locations industry, etc. FMs, e.g., BERT, GPT-4, ViT, serve of data for fine-tuning UFMs.

Agency is a system's capacity to steer outcomes toward a goal, and is a central topic of study across biology, philosophy, cognitive science, and artificial intelligence. Determining if a system exhibits agency is a notoriously difficult question: Dennett (1989), for instance, highlights the puzzle of determining which principles can decide whether a rock, a thermostat, or a robot each possess agency. We here address this puzzle from the viewpoint of reinforcement learning by arguing that agency is fundamentally frame-dependent: Any measurement of a system's agency must be made relative to a reference frame. We support this claim by presenting a philosophical argument that each of the essential properties of agency proposed by Barandiaran et al. (2009) and Moreno (2018) are themselves frame-dependent. We conclude that any basic science of agency requires frame-dependence, and discuss the implications of this claim for reinforcement learning.

Regularization is a core component of recent Reinforcement Learning (RL) algorithms. Mirror Descent Value Iteration (MDVI) uses both Kullback-Leibler divergence and entropy as regularizers in its value and policy updates. Despite its empirical success in discrete action domains and strong theoretical guarantees, the performance of a MDVI-based method does not surpass an entropy-only-regularized method in continuous action domains. In this study, we propose Mirror Descent Actor Critic (MDAC) as an actor-critic style instantiation of MDVI for continuous action domains, and show that its empirical performance is significantly boosted by bounding the actor's log-density terms in the critic's loss function, compared to a non-bounded naive instantiation. Further, we relate MDAC to Advantage Learning by recalling that the actor's log-probability is equal to the regularized advantage function in tabular cases, and theoretically discuss when and why bounding the advantage terms is validated and beneficial. We also empirically explore a good choice for the bounding function, and show that MDAC perfoms better than strong non-regularized and entropy-only-regularized methods with an appropriate choice of the bounding function.

Intelligence, which leverages sociocultural interactions to enhance open-ended skill acquisition. Artificial Intelligence (AI) aims to create autonomous agents that can operate across diverse environments and complete a wide range of tasks. Researchers pursue different approaches, each focusing on specific drivers of learning. In Reinforcement Learning (RL) [1], agents learn by exploring their environment and using their experience to solve tasks. Imitation Learning (IL) [2] involves agents learning from expert demonstrations, while Multi-Agent Reinforcement Learning (MARL) [3] emphasizes cooperation among agents to solve collaborative tasks. Recent advancements in RL have demonstrated success in varied domains, such as playing Atari games [4], mastering chess and Go [5], and controlling stratospheric balloons [6]. IL, combined with transformers [7], has enabled generalist agents to be trained on diverse datasets and to perform in-context reinforcement learning via algorithm distillation. However, these algorithms remain sample-inefficient and struggle with generalization, creativity, and tackling novel tasks, largely because they rely on isolated learning signals. This research explores sociocultural interactions as a new avenue for AI learning inspired by human development.

Reward-guided text generation (RGTG) has emerged as a viable alternative to offline reinforcement learning from human feedback (RLHF). RGTG methods can align baseline language models to human preferences without further training like in standard RLHF methods. However, they rely on a reward model to score each candidate token generated by the language model at inference, incurring significant test-time overhead. Additionally, the reward model is usually only trained to score full sequences, which can lead to sub-optimal choices for partial sequences. In this work, we present a novel reward model architecture that is trained, using a Bradley-Terry loss, to prefer the optimal expansion of a sequence with just a \emph{single call} to the reward model at each step of the generation process. That is, a score for all possible candidate tokens is generated simultaneously, leading to efficient inference. We theoretically analyze various RGTG reward models and demonstrate that prior techniques prefer sub-optimal sequences compared to our method during inference. Empirically, our reward model leads to significantly faster inference than other RGTG methods. It requires fewer calls to the reward model and performs competitively compared to previous RGTG and offline RLHF methods.

Entropy-based objectives are widely used to perform state space exploration in reinforcement learning (RL) and dataset generation for offline RL. Behavioral entropy (BE), a rigorous generalization of classical entropies that incorporates cognitive and perceptual biases of agents, was recently proposed for discrete settings and shown to be a promising metric for robotic exploration problems. In this work, we propose using BE as a principled exploration objective for systematically generating datasets that provide diverse state space coverage in complex, continuous, potentially high-dimensional domains. To achieve this, we extend the notion of BE to continuous settings, derive tractable k-nearest neighbor estimators, provide theoretical guarantees for these estimators, and develop practical reward functions that can be used with standard RL methods to learn BE-maximizing policies. Using standard MuJoCo environments, we experimentally compare the performance of offline RL algorithms for a variety of downstream tasks on datasets generated using BE, Rényi, and Shannon entropy-maximizing policies, as well as the SMM and RND algorithms. We find that offline RL algorithms trained on datasets collected using BE outperform those trained on datasets collected using Shannon entropy, SMM, and RND on all tasks considered, and on 80% of the tasks compared to datasets collected using Rényi entropy. Reinforcement learning (RL) methods can successfully solve challenging tasks in complex environments, even outperforming humans in a variety of cases (Mnih et al., 2015; Silver et al., 2018).