Category Archives: Cognitive Sciences

Improving sample efficiency of RL through memory reconstruction

Y. Kang et al., Sample Efficient Reinforcement Learning Using Graph-Based Memory Reconstruction, IEEE Transactions on Artificial Intelligence, vol. 5, no. 2, pp. 751-762, Feb. 2024 DOI: 10.1109/TAI.2023.3268612.

Reinforcement learning (RL) algorithms typically require orders of magnitude more interactions than humans to learn effective policies. Research on memory in neuroscience suggests that humans’ learning efficiency benefits from associating their experiences and reconstructing potential events. Inspired by this finding, we introduce a human brainlike memory structure for agents and build a general learning framework based on this structure to improve the RL sampling efficiency. Since this framework is similar to the memory reconstruction process in psychology, we name the newly proposed RL framework as graph-based memory reconstruction (GBMR). In particular, GBMR first maintains an attribute graph on the agent’s memory and then retrieves its critical nodes to build and update potential paths among these nodes. This novel pipeline drives the RL agent to learn faster with its memory-enhanced value functions and reduces interactions with the environment by reconstructing its valuable paths. Extensive experimental analyses and evaluations in the grid maze and some challenging Atari environments demonstrate GBMRs superiority over traditional RL methods. We will release the source code and trained models to facilitate further studies in this research direction.

On the complexities of RL when it confronts the real (natural) world

Toby Wise, Kara Emery, Angela Radulescu, Naturalistic reinforcement learning, Trends in Cognitive Sciences, Volume 28, Issue 2, 2024, Pages 144-158 DOI: 10.1016/j.tics.2023.08.016.

Humans possess a remarkable ability to make decisions within real-world environments that are expansive, complex, and multidimensional. Human cognitive computational neuroscience has sought to exploit reinforcement learning (RL) as a framework within which to explain human decision-making, often focusing on constrained, artificial experimental tasks. In this article, we review recent efforts that use naturalistic approaches to determine how humans make decisions in complex environments that better approximate the real world, providing a clearer picture of how humans navigate the challenges posed by real-world decisions. These studies purposely embed elements of naturalistic complexity within experimental paradigms, rather than focusing on simplification, generating insights into the processes that likely underpin humans\u2019 ability to navigate complex, multidimensional real-world environments so successfully.

On the need of interacting with the real world to acquire meaning

Giovanni Pezzulo, Thomas Parr, Paul Cisek, Andy Clark, Karl Friston, Generating meaning: active inference and the scope and limits of passive AI, Trends in Cognitive Sciences, Volume 28, Issue 2, 2024, Pages 97-112, DOI: 10.1016/j.tics.2023.10.002.

Prominent accounts of sentient behavior depict brains as generative models of organismic interaction with the world, evincing intriguing similarities with current advances in generative artificial intelligence (AI). However, because they contend with the control of purposive, life-sustaining sensorimotor interactions, the generative models of living organisms are inextricably anchored to the body and world. Unlike the passive models learned by generative AI systems, they must capture and control the sensory consequences of action. This allows embodied agents to intervene upon their worlds in ways that constantly put their best models to the test, thus providing a solid bedrock that is \u2013 we argue \u2013 essential to the development of genuine understanding. We review the resulting implications and consider future directions for generative AI.

On the relations between symbolic and subsymbolic systems in AI

Giuseppe Marra, Sebastijan Duman\u010di\u0107, Robin Manhaeve, Luc De Raedt, From statistical relational to neurosymbolic artificial intelligence: A survey, Artificial Intelligence, Volume 328, 2024 DOI: 10.1016/j.artint.2023.104062.

This survey explores the integration of learning and reasoning in two different fields of artificial intelligence: neurosymbolic and statistical relational artificial intelligence. Neurosymbolic artificial intelligence (NeSy) studies the integration of symbolic reasoning and neural networks, while statistical relational artificial intelligence (StarAI) focuses on integrating logic with probabilistic graphical models. This survey identifies seven shared dimensions between these two subfields of AI. These dimensions can be used to characterize different NeSy and StarAI systems. They are concerned with (1) the approach to logical inference, whether model or proof-based; (2) the syntax of the used logical theories; (3) the logical semantics of the systems and their extensions to facilitate learning; (4) the scope of learning, encompassing either parameter or structure learning; (5) the presence of symbolic and subsymbolic representations; (6) the degree to which systems capture the original logic, probabilistic, and neural paradigms; and (7) the classes of learning tasks the systems are applied to. By positioning various NeSy and StarAI systems along these dimensions and pointing out similarities and differences between them, this survey contributes fundamental concepts for understanding the integration of learning and reasoning.

Improving sample efficiency in actor-critic RL (A2C with NNs) through multimodal advantage function

Jonghyeok Park, Soohee Han, Reinforcement learning with multimodal advantage function for accurate advantage estimation in robot learning, Engineering Applications of Artificial Intelligence, Volume 126, Part C, 2023 DOI: 10.1016/j.engappai.2023.107019.

In this paper, we propose a reinforcement learning (RL) framework that uses a multimodal advantage function (MAF) to come close to the true advantage function, thereby achieving high returns. The MAF, which is constructed as a logarithm of a mixture of Gaussians policy (MoG-P) and trained by globally collected past experiences, directly assesses the complex true advantage function with its multi-modality and is expected to enhance the sample-efficiency of RL. To realize the expected enhanced learning performance with the proposed RL framework, two practical techniques are developed that include mode selection and rounding off of actions during the policy update process. Mode selection is conducted to sample the action around the most influential or weighted mode for efficient environment exploration. For fast policy updates, past actions are rounded off to discretized action values when calculating the multimodal advantage function. The proposed RL framework was validated using simulation environments and a real inverted pendulum system. The findings showed that the proposed framework can achieve a more sample-efficient performance or higher returns than other advantage-based RL benchmarks.

Learning options in RL and using rewards adequately in that context

Richard S. Sutton, Marlos C. Machado, G. Zacharias Holland, David Szepesvari, Finbarr Timbers, Brian Tanner, Adam White, Reward-respecting subtasks for model-based reinforcement learning, Artificial Intelligence, Volume 324, 2023, DOI: 10.1016/j.artint.2023.104001.

To achieve the ambitious goals of artificial intelligence, reinforcement learning must include planning with a model of the world that is abstract in state and time. Deep learning has made progress with state abstraction, but temporal abstraction has rarely been used, despite extensively developed theory based on the options framework. One reason for this is that the space of possible options is immense, and the methods previously proposed for option discovery do not take into account how the option models will be used in planning. Options are typically discovered by posing subsidiary tasks, such as reaching a bottleneck state or maximizing the cumulative sum of a sensory signal other than reward. Each subtask is solved to produce an option, and then a model of the option is learned and made available to the planning process. In most previous work, the subtasks ignore the reward on the original problem, whereas we propose subtasks that use the original reward plus a bonus based on a feature of the state at the time the option terminates. We show that option models obtained from such reward-respecting subtasks are much more likely to be useful in planning than eigenoptions, shortest path options based on bottleneck states, or reward-respecting options generated by the option-critic. Reward respecting subtasks strongly constrain the space of options and thereby also provide a partial solution to the problem of option discovery. Finally, we show how values, policies, options, and models can all be learned online and off-policy using standard algorithms and general value functions.

Reward machines as reward specification method for RL and their automated learning

Rodrigo Toro Icarte, Toryn Q. Klassen, Richard Valenzano, Margarita P. Castro, Ethan Waldie, Sheila A. McIlraith, Learning reward machines: A study in partially observable reinforcement learning, Artificial Intelligence, Volume 323, 2023 DOI: 10.1016/j.artint.2023.103989.

Reinforcement Learning (RL) is a machine learning paradigm wherein an artificial agent interacts with an environment with the purpose of learning behaviour that maximizes the expected cumulative reward it receives from the environment. Reward machines (RMs) provide a structured, automata-based representation of a reward function that enables an RL agent to decompose an RL problem into structured subproblems that can be efficiently learned via off-policy learning. Here we show that RMs can be learned from experience, instead of being specified by the user, and that the resulting problem decomposition can be used to effectively solve partially observable RL problems. We pose the task of learning RMs as a discrete optimization problem where the objective is to find an RM that decomposes the problem into a set of subproblems such that the combination of their optimal memoryless policies is an optimal policy for the original problem. We show the effectiveness of this approach on three partially observable domains, where it significantly outperforms A3C, PPO, and ACER, and discuss its advantages, limitations, and broader potential.

A review of RL algorithms

Ashish Kumar Shakya, Gopinatha Pillai, Sohom Chakrabarty, Reinforcement learning algorithms: A brief survey, Expert Systems with Applications, Volume 231, 2023 DOI: 10.1016/j.eswa.2023.120495.

Reinforcement Learning (RL) is a machine learning (ML) technique to learn sequential decision-making in complex problems. RL is inspired by trial-and-error based human/animal learning. It can learn an optimal policy autonomously with knowledge obtained by continuous interaction with a stochastic dynamical environment. Problems considered virtually impossible to solve, such as learning to play video games just from pixel information, are now successfully solved using deep reinforcement learning. Without human intervention, RL agents can surpass human performance in challenging tasks. This review gives a broad overview of RL, covering its fundamental principles, essential methods, and illustrative applications. The authors aim to develop an initial reference point for researchers commencing their research work in RL. In this review, the authors cover some fundamental model-free RL algorithms and pathbreaking function approximation-based deep RL (DRL) algorithms for complex uncertain tasks with continuous action and state spaces, making RL useful in various interdisciplinary fields. This article also provides a brief review of model-based and multi-agent RL approaches. Finally, some promising research directions for RL are briefly presented.

They had to do it: Certified RL (through online reward shaping/definition)

Hosein Hasanbeig, Daniel Kroening, Alessandro Abate, Certified reinforcement learning with logic guidance, Artificial Intelligence, Volume 322, 2023 DOI: 10.1016/j.artint.2023.103949.

Reinforcement Learning (RL) is a widely employed machine learning architecture that has been applied to a variety of control problems. However, applications in safety-critical domains require a systematic and formal approach to specifying requirements as tasks or goals. We propose a model-free RL algorithm that enables the use of Linear Temporal Logic (LTL) to formulate a goal for unknown continuous-state/action Markov Decision Processes (MDPs). The given LTL property is translated into a Limit-Deterministic Generalised B�chi Automaton (LDGBA), which is then used to shape a synchronous reward function on-the-fly. Under certain assumptions, the algorithm is guaranteed to synthesise a control policy whose traces satisfy the LTL specification with maximal probability.

Meta-RL: given a distribution of tasks, learn a policy capable of adapting to any new task from the task distribution with as little data as possible

Jacob Beck, Risto Vuorio, Evan Zheran Liu, Zheng Xiong, Luisa Zintgraf, Chelsea Finn, Shimon Whiteson, A Survey of Meta-Reinforcement Learning, arXiv:2301.08028 [cs.LG], 2023 DOI: 10.48550/arXiv.2301.08028.

While deep reinforcement learning (RL) has fueled multiple high-profile successes in machine learning, it is held back from more widespread adoption by its often poor data efficiency and the limited generality of the policies it produces. A promising approach for alleviating these limitations is to cast the development of better RL algorithms as a machine learning problem itself in a process called meta-RL. Meta-RL is most commonly studied in a problem setting where, given a distribution of tasks, the goal is to learn a policy that is capable of adapting to any new task from the task distribution with as little data as possible. In this survey, we describe the meta-RL problem setting in detail as well as its major variations. We discuss how, at a high level, meta-RL research can be clustered based on the presence of a task distribution and the learning budget available for each individual task. Using these clusters, we then survey meta-RL algorithms and applications. We conclude by presenting the open problems on the path to making meta-RL part of the standard toolbox for a deep RL practitioner.