Masashi Okada

Masashi Okada, Tadahiro Taniguchi

The present paper proposes a novel reinforcement learning method with world models, DreamingV2, a collaborative extension of DreamerV2 and Dreaming. DreamerV2 is a cutting-edge model-based reinforcement learning from pixels that uses discrete world models to represent latent states with categorical variables. Dreaming is also a form of reinforcement learning from pixels that attempts to avoid the autoencoding process in general world model training by involving a reconstruction-free contrastive learning objective. The proposed DreamingV2 is a novel approach of adopting both the discrete representation of DreamingV2 and the reconstruction-free objective of Dreaming. Compared to DreamerV2 and other recent model-based methods without reconstruction, DreamingV2 achieves the best scores on five simulated challenging 3D robot arm tasks. We believe that DreamingV2 will be a reliable solution for robot learning since its discrete representation is suitable to describe discontinuous environments, and the reconstruction-free fashion well manages complex vision observations.

Hiroki Nakamura, Masashi Okada, Tadahiro Taniguchi

In this study, a novel self-supervised learning (SSL) method is proposed, which considers SSL in terms of variational inference to learn not only representation but also representation uncertainties. SSL is a method of learning representations without labels by maximizing the similarity between image representations of different augmented views of an image. Meanwhile, variational autoencoder (VAE) is an unsupervised representation learning method that trains a probabilistic generative model with variational inference. Both VAE and SSL can learn representations without labels, but their relationship has not been investigated in the past. Herein, the theoretical relationship between SSL and variational inference has been clarified. Furthermore, a novel method, namely variational inference SimSiam (VI-SimSiam), has been proposed. VI-SimSiam can predict the representation uncertainty by interpreting SimSiam with variational inference and defining the latent space distribution. The present experiments qualitatively show that VI- SimSiam could learn uncertainty by comparing input images and predicted uncertainties. Additionally, we described a relationship between estimated uncertainty and classification accuracy.

Masashi Okada, Tadahiro Taniguchi

This paper deals with a new accelerated path integral method, which iteratively searches optimal controls with a small number of iterations. This study is based on the recent observations that a path integral method for reinforcement learning can be interpreted as gradient descent. This observation also applies to an iterative path integral method for optimal control, which sets a convincing argument for utilizing various optimization methods for gradient descent, such as momentum-based acceleration, step-size adaptation and their combination. We introduce these types of methods to the path integral and demonstrate that momentum-based methods, like Nesterov Accelerated Gradient and Adam, can significantly improve the convergence rate to search for optimal controls in simulated control systems. We also demonstrate that the accelerated path integral could improve the performance on model predictive control for various vehicle navigation tasks. Finally, we represent this accelerated path integral method as a recurrent network, which is the accelerated version of the previously proposed path integral networks (PI-Net). We can train the accelerated PI-Net more efficiently for inverse optimal control with less RAM than the original PI-Net.

Akira Kinose, Masashi Okada, Ryo Okumura et al.

In this paper, we propose Multi-View Dreaming, a novel reinforcement learning agent for integrated recognition and control from multi-view observations by extending Dreaming. Most current reinforcement learning method assumes a single-view observation space, and this imposes limitations on the observed data, such as lack of spatial information and occlusions. This makes obtaining ideal observational information from the environment difficult and is a bottleneck for real-world robotics applications. In this paper, we use contrastive learning to train a shared latent space between different viewpoints, and show how the Products of Experts approach can be used to integrate and control the probability distributions of latent states for multiple viewpoints. We also propose Multi-View DreamingV2, a variant of Multi-View Dreaming that uses a categorical distribution to model the latent state instead of the Gaussian distribution. Experiments show that the proposed method outperforms simple extensions of existing methods in a realistic robot control task.

Hiroki Nakamura, Masashi Okada, Tadahiro Taniguchi

In this paper, we propose a new self-supervised learning (SSL) method for representations that enable logic operations. Representation learning has been applied to various tasks, such as image generation and retrieval. The logical controllability of representations is important for these tasks. Although some methods have been shown to enable the intuitive control of representations using natural languages as the inputs, representation control via logic operations between representations has not been demonstrated. Some SSL methods using representation synthesis (e.g., elementwise mean and maximum operations) have been proposed, but the operations performed in these methods do not incorporate logic operations. In this work, we propose a logic-operable self-supervised representation learning method by replacing the existing representation synthesis with the OR operation on the probabilistic extension of many-valued logic. The representations comprise a set of feature-possession degrees, which are truth values indicating the presence or absence of each feature in the image, and realize the logic operations (e.g., OR and AND). Our method can generate a representation that has the features of both representations or only those features common to both representations. In addition, the expression of the ambiguous presence of a feature is realized by indicating the feature-possession degree by the probability distribution of truth values of the many-valued logic. We showed that our method performs competitively in single and multi-label classification tasks compared with prior SSL methods using synthetic representations. Moreover, experiments on image retrieval using MNIST and PascalVOC showed that the representations of our method can be operated by OR and AND operations.

Hiroki Nakamura, Hiroto Iino, Masashi Okada et al.

We propose EasyControlEdge, adapting an image-generation foundation model to edge detection. In real-world edge detection (e.g., floor-plan walls, satellite roads/buildings, and medical organ boundaries), crispness and data efficiency are crucial, yet producing crisp raw edge maps with limited training samples remains challenging. Although image-generation foundation models perform well on many downstream tasks, their pretrained priors for data-efficient transfer and iterative refinement for high-frequency detail preservation remain underexploited for edge detection. To enable crisp and data-efficient edge detection using these capabilities, we introduce an edge-specialized adaptation of image-generation foundation models. To better specialize the foundation model for edge detection, we incorporate an edge-oriented objective with an efficient pixel-space loss. At inference, we introduce guidance based on unconditional dynamics, enabling a single model to control the edge density through a guidance scale. Experiments on BSDS500, NYUDv2, BIPED, and CubiCasa compare against state-of-the-art methods and show consistent gains, particularly under no-post-processing crispness evaluation and with limited training data.

Toshiyuki T. Yokoyama, Masashi Okada, Tadahiro Taniguchi

Annual recruitment data of new graduates are manually analyzed by human resources specialists (HR) in industries, which signifies the need to evaluate the recruitment strategy of HR specialists. Every year, different applicants send in job applications to companies. The relationships between applicants' attributes (e.g., English skill or academic credential) can be used to analyze the changes in recruitment trends across multiple years' data. However, most attributes are unnormalized and thus require thorough preprocessing. Such unnormalized data hinder the effective comparison of the relationship between applicants in the early stage of data analysis. Thus, a visual exploration system is highly needed to gain insight from the overview of the relationship between applicants across multiple years. In this study, we propose the Polarizing Attributes for Network Analysis of Correlation on Entities Association (Panacea) visualization system. The proposed system integrates a time-varying graph model and dynamic graph visualization for heterogeneous tabular data. Using this system, human resource specialists can interactively inspect the relationships between two attributes of prospective employees across multiple years. Further, we demonstrate the usability of Panacea with representative examples for finding hidden trends in real-world datasets and then describe HR specialists' feedback obtained throughout Panacea's development. The proposed Panacea system enables HR specialists to visually explore the annual recruitment of new graduates.

Masashi Okada, Tadahiro Taniguchi

In the present paper, we propose a decoder-free extension of Dreamer, a leading model-based reinforcement learning (MBRL) method from pixels. Dreamer is a sample- and cost-efficient solution to robot learning, as it is used to train latent state-space models based on a variational autoencoder and to conduct policy optimization by latent trajectory imagination. However, this autoencoding based approach often causes object vanishing, in which the autoencoder fails to perceives key objects for solving control tasks, and thus significantly limiting Dreamer's potential. This work aims to relieve this Dreamer's bottleneck and enhance its performance by means of removing the decoder. For this purpose, we firstly derive a likelihood-free and InfoMax objective of contrastive learning from the evidence lower bound of Dreamer. Secondly, we incorporate two components, (i) independent linear dynamics and (ii) the random crop data augmentation, to the learning scheme so as to improve the training performance. In comparison to Dreamer and other recent model-free reinforcement learning methods, our newly devised Dreamer with InfoMax and without generative decoder (Dreaming) achieves the best scores on 5 difficult simulated robotics tasks, in which Dreamer suffers from object vanishing.

Masashi Okada, Norio Kosaka, Tadahiro Taniguchi

In the present paper, we propose an extension of the Deep Planning Network (PlaNet), also referred to as PlaNet of the Bayesians (PlaNet-Bayes). There has been a growing demand in model predictive control (MPC) in partially observable environments in which complete information is unavailable because of, for example, lack of expensive sensors. PlaNet is a promising solution to realize such latent MPC, as it is used to train state-space models via model-based reinforcement learning (MBRL) and to conduct planning in the latent space. However, recent state-of-the-art strategies mentioned in MBRR literature, such as involving uncertainty into training and planning, have not been considered, significantly suppressing the training performance. The proposed extension is to make PlaNet uncertainty-aware on the basis of Bayesian inference, in which both model and action uncertainty are incorporated. Uncertainty in latent models is represented using a neural network ensemble to approximately infer model posteriors. The ensemble of optimal action candidates is also employed to capture multimodal uncertainty in the optimality. The concept of the action ensemble relies on a general variational inference MPC (VI-MPC) framework and its instance, probabilistic action ensemble with trajectory sampling (PaETS). In this paper, we extend VI-MPC and PaETS, which have been originally introduced in previous literature, to address partially observable cases. We experimentally compare the performances on continuous control tasks, and conclude that our method can consistently improve the asymptotic performance compared with PlaNet.

Ryo Okumura, Masashi Okada, Tadahiro Taniguchi

State representation learning (SRL) in partially observable Markov decision processes has been studied to learn abstract features of data useful for robot control tasks. For SRL, acquiring domain-agnostic states is essential for achieving efficient imitation learning. Without these states, imitation learning is hampered by domain-dependent information useless for control. However, existing methods fail to remove such disturbances from the states when the data from experts and agents show large domain shifts. To overcome this issue, we propose a domain-adversarial and conditional state space model (DAC-SSM) that enables control systems to obtain domain-agnostic and task- and dynamics-aware states. DAC-SSM jointly optimizes the state inference, observation reconstruction, forward dynamics, and reward models. To remove domain-dependent information from the states, the model is trained with domain discriminators in an adversarial manner, and the reconstruction is conditioned on domain labels. We experimentally evaluated the model predictive control performance via imitation learning for continuous control of sparse reward tasks in simulators and compared it with the performance of the existing SRL method. The agents from DAC-SSM achieved performance comparable to experts and more than twice the baselines. We conclude domain-agnostic states are essential for imitation learning that has large domain shifts and can be obtained using DAC-SSM.

Masashi Okada, Shinji Takenaka, Tadahiro Taniguchi

It is an effective strategy for the multi-person pose tracking task in videos to employ prediction and pose matching in a frame-by-frame manner. For this type of approach, uncertainty-aware modeling is essential because precise prediction is impossible. However, previous studies have relied on only a single prediction without incorporating uncertainty, which can cause critical tracking errors if the prediction is unreliable. This paper proposes an extension to this approach with Sequential Monte Carlo (SMC). This naturally reformulates the tracking scheme to handle multiple predictions (or hypotheses) of poses, thereby mitigating the negative effect of prediction errors. An important component of SMC, i.e., a proposal distribution, is designed as a probabilistic neural pose predictor, which can propose diverse and plausible hypotheses by incorporating epistemic uncertainty and heteroscedastic aleatoric uncertainty. In addition, a recurrent architecture is introduced to our neural modeling to utilize time-sequence information of poses to manage difficult situations, such as the frequent disappearance and reappearances of poses. Compared to existing baselines, the proposed method achieves a state-of-the-art MOTA score on the PoseTrack2018 validation dataset by reducing approximately 50% of tracking errors from a state-of-the art baseline method.

Masashi Okada, Tadahiro Taniguchi

In recent studies on model-based reinforcement learning (MBRL), incorporating uncertainty in forward dynamics is a state-of-the-art strategy to enhance learning performance, making MBRLs competitive to cutting-edge model free methods, especially in simulated robotics tasks. Probabilistic ensembles with trajectory sampling (PETS) is a leading type of MBRL, which employs Bayesian inference to dynamics modeling and model predictive control (MPC) with stochastic optimization via the cross entropy method (CEM). In this paper, we propose a novel extension to the uncertainty-aware MBRL. Our main contributions are twofold: Firstly, we introduce a variational inference MPC, which reformulates various stochastic methods, including CEM, in a Bayesian fashion. Secondly, we propose a novel instance of the framework, called probabilistic action ensembles with trajectory sampling (PaETS). As a result, our Bayesian MBRL can involve multimodal uncertainties both in dynamics and optimal trajectories. In comparison to PETS, our method consistently improves asymptotic performance on several challenging locomotion tasks.

Masashi Okada, Luca Rigazio, Takenobu Aoshima

In this paper, we introduce Path Integral Networks (PI-Net), a recurrent network representation of the Path Integral optimal control algorithm. The network includes both system dynamics and cost models, used for optimal control based planning. PI-Net is fully differentiable, learning both dynamics and cost models end-to-end by back-propagation and stochastic gradient descent. Because of this, PI-Net can learn to plan. PI-Net has several advantages: it can generalize to unseen states thanks to planning, it can be applied to continuous control tasks, and it allows for a wide variety learning schemes, including imitation and reinforcement learning. Preliminary experiment results show that PI-Net, trained by imitation learning, can mimic control demonstrations for two simulated problems; a linear system and a pendulum swing-up problem. We also show that PI-Net is able to learn dynamics and cost models latent in the demonstrations.