Yasuo Kuniyoshi

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi

Long-horizon dexterous robot manipulation of deformable objects, such as banana peeling, is a problematic task because of the difficulties in object modeling and a lack of knowledge about stable and dexterous manipulation skills. This paper presents a goal-conditioned dual-action (GC-DA) deep imitation learning (DIL) approach that can learn dexterous manipulation skills using human demonstration data. Previous DIL methods map the current sensory input and reactive action, which often fails because of compounding errors in imitation learning caused by the recurrent computation of actions. The method predicts reactive action only when the precise manipulation of the target object is required (local action) and generates the entire trajectory when precise manipulation is not required (global action). This dual-action formulation effectively prevents compounding error in the imitation learning using the trajectory-based global action while responding to unexpected changes in the target object during the reactive local action. The proposed method was tested in a real dual-arm robot and successfully accomplished the banana-peeling task. Data from this and related works are available at: https://sites.google.com/view/multi-task-fine.

Yoshiyuki Ohmura, Yasuo Kuniyoshi

Why do some physical systems possess consciousness, while others do not? Is this a question of physics? Or is it a question of the theory of causation? Physics and the theory of causation serve different descriptive purposes, and in this study we refer to them respectively as the Physical Stance and the Causal Stance. We propose that the generation of consciousness is determined by its internal causal mechanisms in the Causal Stance. To describe a causal model, we will introduce an asymmetric relation between cause and effect into the formulation that is necessary for describing causality, though not physical laws. We argue that the causal conditions for the generation of consciousness are constituted by internal causal mechanisms of the system, rather than by external interventions. To explain such intrinsic causes, this study focuses on inter-level causality. Traditionally, inter-level causality has been considered an emergent phenomenon rather than a mechanism. We devise a method to implement these mechanisms explicitly in a causal model by examining how causes originating at higher levels are transmitted to lower levels within the system. We then propose a Dual-Laws Model (DLM), which features distinct dynamical laws at higher and lower levels. Finally, we discuss the generation of functional consciousness and its causality based on the DLM. Note that this study does not address the causal efficacy of the phenomenological aspect.

Mitsumasa Nakajima, Katsuma Inoue, Kenji Tanaka et al.

The ever-growing demand for further advances in artificial intelligence motivated research on unconventional computation based on analog physical devices. While such computation devices mimic brain-inspired analog information processing, learning procedures still relies on methods optimized for digital processing such as backpropagation. Here, we present physical deep learning by extending a biologically plausible training algorithm called direct feedback alignment. As the proposed method is based on random projection with arbitrary nonlinear activation, we can train a physical neural network without knowledge about the physical system. In addition, we can emulate and accelerate the computation for this training on a simple and scalable physical system. We demonstrate the proof-of-concept using a hierarchically connected optoelectronic recurrent neural network called deep reservoir computer. By constructing an FPGA-assisted optoelectronic benchtop, we confirmed the potential for accelerated computation with competitive performance on benchmarks. Our results provide practical solutions for the training and acceleration of neuromorphic computation.

Takayuki Komatsu, Yoshiyuki Ohmura, Yasuo Kuniyoshi

Multi-object representation learning aims to represent complex real-world visual input using the composition of multiple objects. Representation learning methods have often used unsupervised learning to segment an input image into individual objects and encode these objects into each latent vector. However, it is not clear how previous methods have achieved the appropriate segmentation of individual objects. Additionally, most of the previous methods regularize the latent vectors using a Variational Autoencoder (VAE). Therefore, it is not clear whether VAE regularization contributes to appropriate object segmentation. To elucidate the mechanism of object segmentation in multi-object representation learning, we conducted an ablation study on MONet, which is a typical method. MONet represents multiple objects using pairs that consist of an attention mask and the latent vector corresponding to the attention mask. Each latent vector is encoded from the input image and attention mask. Then, the component image and attention mask are decoded from each latent vector. The loss function of MONet consists of 1) the sum of reconstruction losses between the input image and decoded component image, 2) the VAE regularization loss of the latent vector, and 3) the reconstruction loss of the attention mask to explicitly encode shape information. We conducted an ablation study on these three loss functions to investigate the effect on segmentation performance. Our results showed that the VAE regularization loss did not affect segmentation performance and the others losses did affect it. Based on this result, we hypothesize that it is important to maximize the attention mask of the image region best represented by a single latent vector corresponding to the attention mask. We confirmed this hypothesis by evaluating a new loss function with the same mechanism as the hypothesis.

Yongbo Zhang, Katsuma Inoue, Mitsumasa Nakajima et al.

Spiking neural networks (SNNs), the models inspired by the mechanisms of real neurons in the brain, transmit and represent information by employing discrete action potentials or spikes. The sparse, asynchronous properties of information processing make SNNs highly energy efficient, leading to SNNs being promising solutions for implementing neural networks in neuromorphic devices. However, the nondifferentiable nature of SNN neurons makes it a challenge to train them. The current training methods of SNNs that are based on error backpropagation (BP) and precisely designing surrogate gradient are difficult to implement and biologically implausible, hindering the implementation of SNNs on neuromorphic devices. Thus, it is important to train SNNs with a method that is both physically implementatable and biologically plausible. In this paper, we propose using augmented direct feedback alignment (aDFA), a gradient-free approach based on random projection, to train SNNs. This method requires only partial information of the forward process during training, so it is easy to implement and biologically plausible. We systematically demonstrate the feasibility of the proposed aDFA-SNNs scheme, propose its effective working range, and analyze its well-performing settings by employing genetic algorithm. We also analyze the impact of crucial features of SNNs on the scheme, thus demonstrating its superiority and stability over BP and conventional direct feedback alignment. Our scheme can achieve competitive performance without accurate prior knowledge about the utilized system, thus providing a valuable reference for physically training SNNs.

JingChuan Guan, Tomoyuki Kubota, Yasuo Kuniyoshi et al.

The effects of noise on memory in a linear recurrent network are theoretically investigated. Memory is characterized by its ability to store previous inputs in its instantaneous state of network, which receives a correlated or uncorrelated noise. Two major properties are revealed: First, the memory reduced by noise is uniquely determined by the noise's power spectral density (PSD). Second, the memory will not decrease regardless of noise intensity if the PSD is in a certain class of distribution (including power law). The results are verified using the human brain signals, showing good agreement.

Alexandre Pitti, Kohei Nakajima, Yasuo Kuniyoshi

The body morphology plays an important role in the way information is perceived and processed by an agent. We address an information theory (IT) account on how the precision of sensors, the accuracy of motors, their placement, the body geometry, shape the information structure in robots and computational codes. As an original idea, we envision the robot's body as a physical communication channel through which information is conveyed, in and out, despite intrinsic noise and material limitations. Following this, entropy, a measure of information and uncertainty, can be used to maximize the efficiency of robot design and of algorithmic codes per se. This is known as the principle of Entropy Maximization (PEM) introduced in biology by Barlow in 1969. The Shannon's source coding theorem provides then a framework to compare different types of bodies in terms of sensorimotor information. In line with PME, we introduce a special class of efficient codes used in IT that reached the Shannon limits in terms of information capacity for error correction and robustness against noise, and parsimony. These efficient codes, which exploit insightfully quantization and randomness, permit to deal with uncertainty, redundancy and compacity. These features can be used for perception and control in intelligent systems. In various examples and closing discussions, we reflect on the broader implications of our framework that we called Informational Embodiment to motor theory and bio-inspired robotics, touching upon concepts like motor synergies, reservoir computing, and morphological computation. These insights can contribute to a deeper understanding of how information theory intersects with the embodiment of intelligence in both natural and artificial systems.

Yoshiyuki Ohmura, Earnest Kota Carr, Yasuo Kuniyoshi

A self-determining system is defined as one in which causes originating within the system influence the system itself. This definition raises the question of how to specify system boundaries. Although the concept of "closure" is commonly used for this purpose, defining boundaries solely in terms of causal relations introduce challenges, such as how to handle external causes and circular causality. To address this issue, we introduce two types of asymmetric relations: causal and constitutive. We propose that system boundaries can be defined as closures of loops formed by these relations, referred to as causal-constitutive loops. By constraining constitutive relations, the resulting system necessarily includes internal causes and thereby satisfies self-determination. Furthermore, to prevent reduction to supervenience, constitutive relations must involve at least two independent variables. This minimal requirement leads to two interdependent loops, which implies a dual-process organization.

Yoshia Abe, Tatsuya Daikoku, Yasuo Kuniyoshi

Recently, it has been recognized that large language models demonstrate high performance on various intellectual tasks. However, few studies have investigated alignment with humans in behaviors that involve sensibility, such as aesthetic evaluation. This study investigates the performance of GPT-4 with Vision, a state-of-the-art language model that can handle image input, on the task of aesthetic evaluation of images. We employ two tasks, prediction of the average evaluation values of a group and an individual's evaluation values. We investigate the performance of GPT-4 with Vision by exploring prompts and analyzing prediction behaviors. Experimental results reveal GPT-4 with Vision's superior performance in predicting aesthetic evaluations and the nature of different responses to beauty and ugliness. Finally, we discuss developing an AI system for aesthetic evaluation based on scientific knowledge of the human perception of beauty, employing agent technologies that integrate traditional deep learning models with large language models.

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi

In the field of robotic manipulation, deep imitation learning is recognized as a promising approach for acquiring manipulation skills. Additionally, learning from diverse robot datasets is considered a viable method to achieve versatility and adaptability. In such research, by learning various tasks, robots achieved generality across multiple objects. However, such multi-task robot datasets have mainly focused on single-arm tasks that are relatively imprecise, not addressing the fine-grained object manipulation that robots are expected to perform in the real world. This paper introduces a dataset of diverse object manipulations that includes dual-arm tasks and/or tasks requiring fine manipulation. To this end, we have generated dataset with 224k episodes (150 hours, 1,104 language instructions) which includes dual-arm fine tasks such as bowl-moving, pencil-case opening or banana-peeling, and this data is publicly available. Additionally, this dataset includes visual attention signals as well as dual-action labels, a signal that separates actions into a robust reaching trajectory and precise interaction with objects, and language instructions to achieve robust and precise object manipulation. We applied the dataset to our Dual-Action and Attention (DAA), a model designed for fine-grained dual arm manipulation tasks and robust against covariate shifts. The model was tested with over 7k total trials in real robot manipulation tasks, demonstrating its capability in fine manipulation.

Yoshiyuki Ohmura, Wataru Shimaya, Yasuo Kuniyoshi

In general, objects can be distinguished on the basis of their features, such as color or shape. In particular, it is assumed that similarity judgments about such features can be processed independently in different metric spaces. However, the unsupervised categorization mechanism of metric spaces corresponding to object features remains unknown. Here, we show that the artificial neural network system can autonomously categorize metric spaces through representation learning to satisfy the algebraic independence between neural networks, and project sensory information onto multiple high-dimensional metric spaces to independently evaluate the differences and similarities between features. Conventional methods often constrain the axes of the latent space to be mutually independent or orthogonal. However, the independent axes are not suitable for categorizing metric spaces. High-dimensional metric spaces that are independent of each other are not uniquely determined by the mutually independent axes, because any combination of independent axes can form mutually independent spaces. In other words, the mutually independent axes cannot be used to naturally categorize different feature spaces, such as color space and shape space. Therefore, constraining the axes to be mutually independent makes it difficult to categorize high-dimensional metric spaces. To overcome this problem, we developed a method to constrain only the spaces to be mutually independent and not the composed axes to be independent. Our theory provides general conditions for the unsupervised categorization of independent metric spaces, thus advancing the mathematical theory of functional differentiation of neural networks.

Kayato Nishitsunoi, Yoshiyuki Ohmura, Takayuki Komatsu et al.

Humans develop certain cognitive abilities to recognize objects and their transformations without explicit supervision, highlighting the importance of unsupervised representation learning. A fundamental challenge in unsupervised representation learning is to separate different transformations in learned feature representations. Although algebraic approaches have been explored, a comprehensive theoretical framework remains underdeveloped. Existing methods decompose transformations based on algebraic independence, but these methods primarily focus on commutative transformations and do not extend to cases where transformations are conditionally independent but noncommutative. To extend current representation learning frameworks, we draw inspiration from Galois theory, where the decomposition of groups through normal subgroups provides an approach for the analysis of structured transformations. Normal subgroups naturally extend commutativity under certain conditions and offer a foundation for the categorization of transformations, even when they do not commute. In this paper, we propose a novel approach that leverages normal subgroups to enable the separation of conditionally independent transformations, even in the absence of commutativity. Through experiments on geometric transformations in images, we show that our method successfully categorizes conditionally independent transformations, such as rotation and translation, in an unsupervised manner, suggesting a close link between group decomposition via normal subgroups and transformation categorization in representation learning.

JingChuan Guan, Tomoyuki Kubota, Yasuo Kuniyoshi et al.

State space models (SSMs) have gained attention by showing potential to outperform Transformers. However, previous studies have not sufficiently addressed the mechanisms underlying their high performance owing to a lack of theoretical explanation of SSMs' learning dynamics. In this study, we provide such an explanation and propose an improved training strategy. The memory capacity of SSMs can be evaluated by examining how input time series are stored in their current state. Such an examination reveals a tradeoff between memory accuracy and length, as well as the theoretical equivalence between the structured state space sequence model (S4) and a simplified S4 with diagonal recurrent weights. This theoretical foundation allows us to elucidate the learning dynamics, proving the importance of initial parameters. Our analytical results suggest that successful learning requires the initial memory structure to be the longest possible even if memory accuracy may deteriorate or the gradient lose the teacher information. Experiments on tasks requiring long memory confirmed that extending memory is difficult, emphasizing the importance of initialization. Furthermore, we found that fixing recurrent weights can be more advantageous than adapting them because it achieves comparable or even higher performance with faster convergence. Our results provide a new theoretical foundation for SSMs and potentially offer a novel optimization strategy.

Takeshi Kobayashi, Shogo Yonekura, Yasuo Kuniyoshi

Neuronal spikes directly drive muscles and endow animals with agile movements, but applying the spike-based control signals to actuators in artificial sensor-motor systems inevitably causes a collapse of learning. We developed a system that can vary \emph{the number of independent synaptic bundles} in sensor-to-motor connections. This paper demonstrates the following four findings: (i) Learning collapses once the number of motor neurons or the number of independent synaptic bundles exceeds a critical limit. (ii) The probability of learning failure is increased by a smaller number of motor neurons, while (iii) if learning succeeds, a smaller number of motor neurons leads to faster learning. (iv) The number of weight updates that move in the opposite direction of the optimal weight can quantitatively explain these results. The functions of spikes remain largely unknown. Identifying the parameter range in which learning systems using spikes can be constructed will make it possible to study the functions of spikes that were previously inaccessible due to the difficulty of learning.

Takayuki Komatsu, Yoshiyuki Ohmura, Kayato Nishitsunoi et al.

The main goal of representation learning is to acquire meaningful representations from real-world sensory inputs without supervision. Representation learning explains some aspects of human development. Various neural network (NN) models have been proposed that acquire empirically good representations. However, the formulation of a good representation has not been established. We recently proposed a method for categorizing changes between a pair of sensory inputs. A unique feature of this approach is that transformations between two sensory inputs are learned to satisfy algebraic structural constraints. Conventional representation learning often assumes that disentangled independent feature axes is a good representation; however, we found that such a representation cannot account for conditional independence. To overcome this problem, we proposed a new method using group decomposition in Galois algebra theory. Although this method is promising for defining a more general representation, it assumes pixel-to-pixel translation without feature extraction, and can only process low-resolution images with no background, which prevents real-world application. In this study, we provide a simple method to apply our group decomposition theory to a more realistic scenario by combining feature extraction and object segmentation. We replace pixel translation with feature translation and formulate object segmentation as grouping features under the same transformation. We validated the proposed method on a practical dataset containing both real-world object and background. We believe that our model will lead to a better understanding of human development of object recognition in the real world.

Pengcheng Pan, Yonekura Shogo, Yasuo Kuniyoshi

Inspired by foveal vision, hard attention models promise interpretability and parameter economy. However, existing models like the Recurrent Model of Visual Attention (RAM) and Deep Recurrent Attention Model (DRAM) failed to model the hierarchy of human vision system, that compromise on the visual exploration dynamics. As a result, they tend to produce attention that are either overly fixational or excessively saccadic, diverging from human eye movement behavior. In this paper, we propose a Multi-Level Recurrent Attention Model (MRAM), a novel hard attention framework that explicitly models the neural hierarchy of human visual processing. By decoupling the function of glimpse location generation and task execution in two recurrent layers, MRAM emergent a balanced behavior between fixation and saccadic movement. Our results show that MRAM not only achieves more human-like attention dynamics, but also consistently outperforms CNN, RAM and DRAM baselines on standard image classification benchmarks.

Tempei Kabayama, Motomasa Komuro, Yasuo Kuniyoshi et al.

Reservoir computing can embed attractors into random neural networks (RNNs), generating a ``mirror'' of a target attractor because of its inherent symmetrical constraints. In these RNNs, we report that an attractor-merging crisis accompanied by intermittency emerges simply by adjusting the global parameter. We further reveal its underlying mechanism through a detailed analysis of the phase-space structure and demonstrate that this bifurcation scenario is intrinsic to a general class of RNNs, independent of training data.

Tempei Kabayama, Yasuo Kuniyoshi, Kazuyuki Aihara et al.

Chaotic dynamics are ubiquitous in nature and useful in engineering, but their geometric design can be challenging. Here, we propose a method using reservoir computing to generate chaos with a desired shape by providing a periodic orbit as a template, called a skeleton. We exploit a bifurcation of the reservoir to intentionally induce unsuccessful training of the skeleton, revealing inherent chaos. The emergence of this untrained attractor, resulting from the interaction between the skeleton and the reservoir's intrinsic dynamics, offers a novel semi-supervised framework for designing chaos.

Heecheol Kim, Yoshiyuki Ohmura, Akihiko Nagakubo et al.

Deep imitation learning is promising for robot manipulation because it only requires demonstration samples. In this study, deep imitation learning is applied to tasks that require force feedback. However, existing demonstration methods have deficiencies; bilateral teleoperation requires a complex control scheme and is expensive, and kinesthetic teaching suffers from visual distractions from human intervention. This research proposes a new master-to-robot (M2R) policy transfer system that does not require robots for teaching force feedback-based manipulation tasks. The human directly demonstrates a task using a controller. This controller resembles the kinematic parameters of the robot arm and uses the same end-effector with force/torque (F/T) sensors to measure the force feedback. Using this controller, the operator can feel force feedback without a bilateral system. The proposed method can overcome domain gaps between the master and robot using gaze-based imitation learning and a simple calibration method. Furthermore, a Transformer is applied to infer policy from F/T sensory input. The proposed system was evaluated on a bottle-cap-opening task that requires force feedback.

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi

Deep imitation learning is a promising approach that does not require hard-coded control rules in autonomous robot manipulation. The current applications of deep imitation learning to robot manipulation have been limited to reactive control based on the states at the current time step. However, future robots will also be required to solve tasks utilizing their memory obtained by experience in complicated environments (e.g., when the robot is asked to find a previously used object on a shelf). In such a situation, simple deep imitation learning may fail because of distractions caused by complicated environments. We propose that gaze prediction from sequential visual input enables the robot to perform a manipulation task that requires memory. The proposed algorithm uses a Transformer-based self-attention architecture for the gaze estimation based on sequential data to implement memory. The proposed method was evaluated with a real robot multi-object manipulation task that requires memory of the previous states.

Takayuki Kanai, Yoshiyuki Ohmura, Akihiko Nagakubo et al.

A robotic hand design suitable for dexterity should be examined using functional tests. To achieve this, we designed a mechanical glove, which is a rigid wearable glove that enables us to develop the corresponding isomorphic robotic hand and evaluate its hardware properties. Subsequently, the effectiveness of multiple degrees-of-freedom (DOFs) was evaluated by human participants. Several fine motor skills were evaluated using the mechanical glove under two conditions: one- and three-DOF conditions. To the best of our knowledge, this is the first extensive evaluation method for robotic hand designs suitable for dexterity. (This paper was peer-reviewed and is the full translation from the Journal of the Robotics Society of Japan, Vol.39, No.6, pp.557-560, 2021.)

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi

Deep imitation learning is promising for solving dexterous manipulation tasks because it does not require an environment model and pre-programmed robot behavior. However, its application to dual-arm manipulation tasks remains challenging. In a dual-arm manipulation setup, the increased number of state dimensions caused by the additional robot manipulators causes distractions and results in poor performance of the neural networks. We address this issue using a self-attention mechanism that computes dependencies between elements in a sequential input and focuses on important elements. A Transformer, a variant of self-attention architecture, is applied to deep imitation learning to solve dual-arm manipulation tasks in the real world. The proposed method has been tested on dual-arm manipulation tasks using a real robot. The experimental results demonstrated that the Transformer-based deep imitation learning architecture can attend to the important features among the sensory inputs, therefore reducing distractions and improving manipulation performance when compared with the baseline architecture without the self-attention mechanisms. Data from this and related works are available at: https://sites.google.com/view/multi-task-fine.

Ryoya Ogishima, Izumi Karino, Yasuo Kuniyoshi

Reinforcement Learning (RL) requires a large amount of exploration especially in sparse-reward settings. Imitation Learning (IL) can learn from expert demonstrations without exploration, but it never exceeds the expert's performance and is also vulnerable to distributional shift between demonstration and execution. In this paper, we radically unify RL and IL based on Free Energy Principle (FEP). FEP is a unified Bayesian theory of the brain that explains perception, action and model learning by a common fundamental principle. We present a theoretical extension of FEP and derive an algorithm in which an agent learns the world model that internalizes expert demonstrations and at the same time uses the model to infer the current and future states and actions that maximize rewards. The algorithm thus reduces exploration costs by partially imitating experts as well as maximizing its return in a seamless way, resulting in a higher performance than the suboptimal expert. Our experimental results show that this approach is promising in visual control tasks especially in sparse-reward environments.

Katsuma Inoue, Soh Ohara, Yasuo Kuniyoshi et al.

Language is an outcome of our complex and dynamic human-interactions and the technique of natural language processing (NLP) is hence built on human linguistic activities. Bidirectional Encoder Representations from Transformers (BERT) has recently gained its popularity by establishing the state-of-the-art scores in several NLP benchmarks. A Lite BERT (ALBERT) is literally characterized as a lightweight version of BERT, in which the number of BERT parameters is reduced by repeatedly applying the same neural network called Transformer's encoder layer. By pre-training the parameters with a massive amount of natural language data, ALBERT can convert input sentences into versatile high-dimensional vectors potentially capable of solving multiple NLP tasks. In that sense, ALBERT can be regarded as a well-designed high-dimensional dynamical system whose operator is the Transformer's encoder, and essential structures of human language are thus expected to be encapsulated in its dynamics. In this study, we investigated the embedded properties of ALBERT to reveal how NLP tasks are effectively solved by exploiting its dynamics. We thereby aimed to explore the nature of human language from the dynamical expressions of the NLP model. Our short-term analysis clarified that the pre-trained model stably yields trajectories with higher dimensionality, which would enhance the expressive capacity required for NLP tasks. Also, our long-term analysis revealed that ALBERT intrinsically shows transient chaos, a typical nonlinear phenomenon showing chaotic dynamics only in its transient, and the pre-trained ALBERT model tends to produce the chaotic trajectory for a significantly longer time period compared to a randomly-initialized one. Our results imply that local chaoticity would contribute to improving NLP performance, uncovering a novel aspect in the role of chaotic dynamics in human language behaviors.

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi

A high-precision manipulation task, such as needle threading, is challenging. Physiological studies have proposed connecting low-resolution peripheral vision and fast movement to transport the hand into the vicinity of an object, and using high-resolution foveated vision to achieve the accurate homing of the hand to the object. The results of this study demonstrate that a deep imitation learning based method, inspired by the gaze-based dual resolution visuomotor control system in humans, can solve the needle threading task. First, we recorded the gaze movements of a human operator who was teleoperating a robot. Then, we used only a high-resolution image around the gaze to precisely control the thread position when it was close to the target. We used a low-resolution peripheral image to reach the vicinity of the target. The experimental results obtained in this study demonstrate that the proposed method enables precise manipulation tasks using a general-purpose robot manipulator and improves computational efficiency. Data from this and related works are available at: https://sites.google.com/view/multi-task-fine.

Izumi Karino, Yoshiyuki Ohmura, Yasuo Kuniyoshi

The balance of exploration and exploitation plays a crucial role in accelerating reinforcement learning (RL). To deploy an RL agent in human society, its explainability is also essential. However, basic RL approaches have difficulties in deciding when to choose exploitation as well as in extracting useful points for a brief explanation of its operation. One reason for the difficulties is that these approaches treat all states the same way. Here, we show that identifying critical states and treating them specially is commonly beneficial to both problems. These critical states are the states at which the action selection changes the potential of success and failure substantially. We propose to identify the critical states using the variance in the Q-function for the actions and to perform exploitation with high probability on the identified states. These simple methods accelerate RL in a grid world with cliffs and two baseline tasks of deep RL. Our results also demonstrate that the identified critical states are intuitively interpretable regarding the crucial nature of the action selection. Furthermore, our analysis of the relationship between the timing of the identification of especially critical states and the rapid progress of learning suggests there are a few especially critical states that have important information for accelerating RL rapidly.

Katsuma Inoue, Kohei Nakajima, Yasuo Kuniyoshi

Chaotic itinerancy is a frequently observed phenomenon in high-dimensional and nonlinear dynamical systems, and it is characterized by the random transitions among multiple quasi-attractors. Several studies have revealed that chaotic itinerancy has been observed in brain activity, and it is considered to play a critical role in the spontaneous, stable behavior generation of animals. Thus, chaotic itinerancy is a topic of great interest, particularly for neurorobotics researchers who wish to understand and implement autonomous behavioral controls for agents. However, it is generally difficult to gain control over high-dimensional nonlinear dynamical systems. Hence, the implementation of chaotic itinerancy has mainly been accomplished heuristically. In this study, we propose a novel way of implementing chaotic itinerancy reproducibly and at will in a generic high-dimensional chaotic system. In particular, we demonstrate that our method enables us to easily design both the trajectories of quasi-attractors and the transition rules among them simply by adjusting the limited number of system parameters and by utilizing the intrinsic high-dimensional chaos. Finally, we quantitatively discuss the validity and scope of application through the results of several numerical experiments.

Izumi Karino, Kazutoshi Tanaka, Ryuma Niiyama et al.

This paper proposes an exploration method for deep reinforcement learning based on parameter space noise. Recent studies have experimentally shown that parameter space noise results in better exploration than the commonly used action space noise. Previous methods devised a way to update the diagonal covariance matrix of a noise distribution and did not consider the direction of the noise vector and its correlation. In addition, fast updates of the noise distribution are required to facilitate policy learning. We propose a method that deforms the noise distribution according to the accumulated returns and the noises that have led to the returns. Moreover, this method switches isotropic exploration and directional exploration in parameter space with regard to obtained rewards. We validate our exploration strategy in the OpenAI Gym continuous environments and modified environments with sparse rewards. The proposed method achieves results that are competitive with a previous method at baseline tasks. Moreover, our approach exhibits better performance in sparse reward environments by exploration with the switching strategy.