Neural networks are vulnerable to input perturbations such as additive noise and adversarial attacks. In contrast, human perception is much more robust to such perturbations. The Bayesian brain hypothesis states that human brains use an internal generative model to update the posterior beliefs of the sensory input. This mechanism can be interpreted as a form of self-consistency between the maximum a posteriori (MAP) estimation of an internal generative model and the external environment. Inspired by such hypothesis, we enforce self-consistency in neural networks by incorporating generative recurrent feedback. We instantiate this design on convolutional neural networks (CNNs). The proposed framework, termed Convolutional Neural Networks with Feedback (CNN-F), introduces a generative feedback with latent variables to existing CNN architectures, where consistent predictions are made through alternating MAP inference under a Bayesian framework. In the experiments, CNN-F shows considerably improved adversarial robustness over conventional feedforward CNNs on standard benchmarks.
Qualcomm AI Research, Qualcomm Technologies Netherlands B.V.（高通人工智能研究院，高通技术荷兰有限公司）；Qualcomm AI Research, Qualcomm Technologies, Inc.（高通人工智能研究院，高通技术公司）；University of Oxford（牛津大学）
Recent advances in deep generative modeling have enabled efficient modeling of high dimensional data distributions and opened up a new horizon for solving data compression problems. Specifically, autoencoder based learned image or video compression solutions are emerging as strong competitors to traditional approaches. In this work, We propose a new network architecture, based on common and well studied components, for learned video compression operating in low latency mode. Our method yields state of the art MS-SSIM/rate performance on the high-resolution UVG dataset, among both learned video…
Sensory robustness through top-down feedback and neural stochasticity in recurrent vision models
Hertie Institute for Clinical Brain Research, University of Tübingen（图宾根大学赫蒂临床脑研究所）；University of Tübingen（图宾根大学）；Center for Bionic Intelligence Tübingen Stuttgart (BITS)（图宾根-斯图加特仿生智能中心）；Institute of Cognitive Sciences and Technologies, National Research Council（国家研究委员会认知科学与技术研究所）；Institute of Neurology, University College London（伦敦大学学院神经病学研究所）；German Center for Mental Health (DZPG)（德国心理健康中心）
Generative adversarial networks have been proposed as a way of efficiently training deep generative neural networks. We propose a generative adversarial model that works on continuous sequential data, and apply it by training it on a collection of classical music. We conclude that it generates music that sounds better and better as the model is trained, report statistics on generated music, and let the reader judge the quality by downloading the generated songs.
Montreal Institute for Learning Algorithms（蒙特利尔学习算法研究所）；University of Montreal（蒙特利尔大学）；York University（约克大学）
Gatys et al. (2015) showed that optimizing pixels to match features in a convolutional network with respect reference image features is a way to render images of high visual quality. We show that unrolling this gradient-based optimization yields a recurrent computation that creates images by incrementally adding onto a visual "canvas". We propose a recurrent generative model inspired by this view, and show that it can be trained using adversarial training to generate very good image samples. We also propose a way to quantitatively compare adversarial networks by having the generators and discr…
Effective defense of deep neural networks against adversarial attacks remains a challenging problem, especially under powerful white-box attacks. In this paper, we develop a new method called ensemble generative cleaning with feedback loops (EGC-FL) for effective defense of deep neural networks. The proposed EGC-FL method is based on two central ideas. First, we introduce a transformed deadzone layer into the defense network, which consists of an orthonormal transform and a deadzone-based activation function, to destroy the sophisticated noise pattern of adversarial attacks. Second, by constru…
We begin by reiterating that common neural network activation functions have simple Bayesian origins. In this spirit, we go on to show that Bayes's theorem also implies a simple recurrence relation; this leads to a Bayesian recurrent unit with a prescribed feedback formulation. We show that introduction of a context indicator leads to a variable feedback that is similar to the forget mechanism in conventional recurrent units. A similar approach leads to a probabilistic input gate. The Bayesian formulation leads naturally to the two pass algorithm of the Kalman smoother or forward-backward algo…
In this work, we propose a novel recurrent neural network (RNN) architecture. The proposed RNN, gated-feedback RNN (GF-RNN), extends the existing approach of stacking multiple recurrent layers by allowing and controlling signals flowing from upper recurrent layers to lower layers using a global gating unit for each pair of layers. The recurrent signals exchanged between layers are gated adaptively based on the previous hidden states and the current input. We evaluated the proposed GF-RNN with different types of recurrent units, such as tanh, long short-term memory and gated recurrent units, on…
Recurrent neural nets are widely used for predicting temporal data. Their inherent deep feedforward structure allows learning complex sequential patterns. It is believed that top-down feedback might be an important missing ingredient which in theory could help disambiguate similar patterns depending on broader context. In this paper we introduce surprisal-driven recurrent networks, which take into account past error information when making new predictions. This is achieved by continuously monitoring the discrepancy between most recent predictions and the actual observations. Furthermore, we sh…
Mitsubishi Electric Research Laboratories (MERL)（三菱电机研究实验室）；Boston University（波士顿大学）
We propose a novel {\it Equilibrated Recurrent Neural Network} (ERNN) to combat the issues of inaccuracy and instability in conventional RNNs. Drawing upon the concept of autapse in neuroscience, we propose augmenting an RNN with a time-delayed self-feedback loop. Our sole purpose is to modify the dynamics of each internal RNN state and, at any time, enforce it to evolve close to the equilibrium point associated with the input signal at that time. We show that such self-feedback helps stabilize the hidden state transitions leading to fast convergence during training while efficiently learning …
Frankfurt Institute for Advanced Studies（法兰克福高等研究院）；Goethe-Universität Frankfurt（法兰克福大学）
Recurrent connectivity in the visual cortex is believed to aid object recognition for challenging conditions such as occlusion. Here we investigate if and how artificial neural networks also benefit from recurrence. We compare architectures composed of bottom-up, lateral and top-down connections and evaluate their performance using two novel stereoscopic occluded object datasets. We find that classification accuracy is significantly higher for recurrent models when compared to feedforward models of matched parametric complexity. Additionally we show that for challenging stimuli, the recurrent …
University of Science and Technology of China（中国科学技术大学）；York University（约克大学）；IFLYTEK Research（科大讯飞研究院）
We introduce a new structure for memory neural networks, called feedforward sequential memory networks (FSMN), which can learn long-term dependency without using recurrent feedback. The proposed FSMN is a standard feedforward neural networks equipped with learnable sequential memory blocks in the hidden layers. In this work, we have applied FSMN to several language modeling (LM) tasks. Experimental results have shown that the memory blocks in FSMN can learn effective representations of long history. Experiments have shown that FSMN based language models can significantly outperform not only fe…
We empirically characterize the performance of discriminative and generative LSTM models for text classification. We find that although RNN-based generative models are more powerful than their bag-of-words ancestors (e.g., they account for conditional dependencies across words in a document), they have higher asymptotic error rates than discriminatively trained RNN models. However we also find that generative models approach their asymptotic error rate more rapidly than their discriminative counterparts---the same pattern that Ng & Jordan (2001) proved holds for linear classification models th…
Qualcomm AI Research, Qualcomm Technologies, Inc.（高通技术公司高通人工智能研究院）；Qualcomm AI Research, Qualcomm Technologies Netherlands B.V.（高通技术荷兰私人有限公司高通人工智能研究院）；University of California San Diego（加州大学圣地亚哥分校）
In this work, we propose a new recurrent autoencoder architecture, termed Feedback Recurrent AutoEncoder (FRAE), for online compression of sequential data with temporal dependency. The recurrent structure of FRAE is designed to efficiently extract the redundancy along the time dimension and allows a compact discrete representation of the data to be learned. We demonstrate its effectiveness in speech spectrogram compression. Specifically, we show that the FRAE, paired with a powerful neural vocoder, can produce high-quality speech waveforms at a low, fixed bitrate. We further show that by addin…
Korea Advanced Institute of Science and Technology (KAIST)（韩国科学技术院（KAIST））
Throughout this paper, we focus on the improvement of the direct feedback alignment (DFA) algorithm and extend the usage of the DFA to convolutional and recurrent neural networks (CNNs and RNNs). Even though the DFA algorithm is biologically plausible and has a potential of high-speed training, it has not been considered as the substitute for back-propagation (BP) due to the low accuracy in the CNN and RNN training. In this work, we propose a new DFA algorithm for BP-level accurate CNN and RNN training. Firstly, we divide the network into several modules and apply the DFA algorithm within the …
The abundant recurrent horizontal and feedback connections in the primate visual cortex are thought to play an important role in bringing global and semantic contextual information to early visual areas during perceptual inference, helping to resolve local ambiguity and fill in missing details. In this study, we find that introducing feedback loops and horizontal recurrent connections to a deep convolution neural network (VGG16) allows the network to become more robust against noise and occlusion during inference, even in the initial feedforward pass. This suggests that recurrent feedback and …
Feedback-driven recurrent spiking neural networks (RSNNs) are powerful computational models that can mimic dynamical systems. However, the presence of a feedback loop from the readout to the recurrent layer de-stabilizes the learning mechanism and prevents it from converging. Here, we propose a supervised training procedure for RSNNs, where a second network is introduced only during the training, to provide hint for the target dynamics. The proposed training procedure consists of generating targets for both recurrent and readout layers (i.e., for a full RSNN system). It uses the recursive leas…
Reliable 4D aircraft trajectory prediction, whether in a real-time setting or for analysis of counterfactuals, is important to the efficiency of the aviation system. Toward this end, we first propose a highly generalizable efficient tree-based matching algorithm to construct image-like feature maps from high-fidelity meteorological datasets - wind, temperature and convective weather. We then model the track points on trajectories as conditional Gaussian mixtures with parameters to be learned from our proposed deep generative model, which is an end-to-end convolutional recurrent neural network …
Tel Aviv University（特拉维夫大学）；Facebook AI Research（脸书人工智能研究院）
The current leading computer vision models are typically feed forward neural models, in which the output of one computational block is passed to the next one sequentially. This is in sharp contrast to the organization of the primate visual cortex, in which feedback and lateral connections are abundant. In this work, we propose a computational model for the role of lateral connections in a given block, in which the weights of the block vary dynamically as a function of its activations, and the input from the upstream blocks is iteratively reintroduced. We demonstrate how this novel architectura…
Modeling the distribution of natural images is challenging, partly because of strong statistical dependencies which can extend over hundreds of pixels. Recurrent neural networks have been successful in capturing long-range dependencies in a number of problems but only recently have found their way into generative image models. We here introduce a recurrent image model based on multi-dimensional long short-term memory units which are particularly suited for image modeling due to their spatial structure. Our model scales to images of arbitrary size and its likelihood is computationally tractable…
School of Computer Science University of St. Gallen（圣加尔登大学计算机科学学院）；Donostia International Physics Center（多尼亚国际物理中心）；Department of Mathematics, Imperial College（帝国理工学院数学系）；School of Physical and Mathematical Sciences Nanyang Technological University（南洋理工大学物理与数学科学学院）
In multiple-input multiple-output (MIMO) systems, it is crucial of utilizing the available channel state information (CSI) at the transmitter for precoding to improve the performance of frequency division duplex (FDD) networks. One of the mainchallenges is to compress a large amount of CSI in CSI feedback transmission in massive MIMO systems. In this paper, we propose a deep learning (DL)-based approach that uses a deep recurrent neural network (RNN) to learn temporal correlation and adopts depthwise separable convolution to shrink the model. The feature extraction module is also elaborately d…
Yale University（耶鲁大学）；Yale School of Medicine（耶鲁医学院）
Recurrent neural networks (RNNs) were designed for dealing with time-series data and have recently been used for creating predictive models from functional magnetic resonance imaging (fMRI) data. However, gathering large fMRI datasets for learning is a difficult task. Furthermore, network interpretability is unclear. To address these issues, we utilize multitask learning and design a novel RNN-based model that learns to discriminate between classes while simultaneously learning to generate the fMRI time-series data. Employing the long short-term memory (LSTM) structure, we develop a discrimina…
This paper presents a generative approach to speech enhancement based on a recurrent variational autoencoder (RVAE). The deep generative speech model is trained using clean speech signals only, and it is combined with a nonnegative matrix factorization noise model for speech enhancement. We propose a variational expectation-maximization algorithm where the encoder of the RVAE is fine-tuned at test time, to approximate the distribution of the latent variables given the noisy speech observations. Compared with previous approaches based on feed-forward fully-connected architectures, the proposed …
Graduate School of Culture Technology, KAIST（韩国科学技术院文化技术研究生院）
End-to-end learning models using raw waveforms as input have shown superior performances in many audio recognition tasks. However, most model architectures are based on convolutional neural networks (CNN) which were mainly developed for visual recognition tasks. In this paper, we propose an extension of squeeze-and-excitation networks (SENets) which adds temporal feedback control from the top-layer features to channel-wise feature activations in lower layers using a recurrent module. This is analogous to the adaptive gain control mechanism of outer hair-cell in the human auditory system. We ap…
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Adobe Research（Adobe 研究院）
The success of various applications including robotics, digital content creation, and visualization demand a structured and abstract representation of the 3D world from limited sensor data. Inspired by the nature of human perception of 3D shapes as a collection of simple parts, we explore such an abstract shape representation based on primitives. Given a single depth image of an object, we present 3D-PRNN, a generative recurrent neural network that synthesizes multiple plausible shapes composed of a set of primitives. Our generative model encodes symmetry characteristics of common man-made obj…
Bidirectional recurrent neural networks (RNN) are trained to predict both in the positive and negative time directions simultaneously. They have not been used commonly in unsupervised tasks, because a probabilistic interpretation of the model has been difficult. Recently, two different frameworks, GSN and NADE, provide a connection between reconstruction and probabilistic modeling, which makes the interpretation possible. As far as we know, neither GSN or NADE have been studied in the context of time series before. As an example of an unsupervised task, we study the problem of filling in gaps …
Singapore Management University（新加坡管理大学）；Indian Institute of Technology Kanpur（印度理工学院坎普尔分校）
Realistic fine-grained multi-agent simulation of real-world complex systems is crucial for many downstream tasks such as reinforcement learning. Recent work has used generative models (GANs in particular) for providing high-fidelity simulation of real-world systems. However, such generative models are often monolithic and miss out on modeling the interaction in multi-agent systems. In this work, we take a first step towards building multiple interacting generative models (GANs) that reflects the interaction in real world. We build and analyze a hierarchical set-up where a higher-level GAN is c…
Institut national de la recherche scientifique (INRS-EMT)（国家科学研究院（INRS-EMT））
In this work, we introduce a two-step framework for generative modeling of temporal data. Specifically, the generative adversarial networks (GANs) setting is employed to generate synthetic scenes of moving objects. To do so, we propose a two-step training scheme within which: a generator of static frames is trained first. Afterwards, a recurrent model is trained with the goal of providing a sequence of inputs to the previously trained frames generator, thus yielding scenes which look natural. The adversarial setting is employed in both training steps. However, with the aim of avoiding known tr…
This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of complex images. The system substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.
Generative System Dynamics in Recurrent Neural Networks
Department of Information Engineering and Mathematics, University of Siena（信息工程与数学系，锡耶纳大学）；Department of Information Engineering, University of Florence（信息工程系，佛罗伦萨大学）；IMT School for Advanced Studies Lucca（卢塞恩高级研究学院）
The effectiveness of biosignal generation and data augmentation with biosignal generative models based on generative adversarial networks (GANs), which are a type of deep learning technique, was demonstrated in our previous paper. GAN-based generative models only learn the projection between a random distribution as input data and the distribution of training data.Therefore, the relationship between input and generated data is unclear, and the characteristics of the data generated from this model cannot be controlled. This study proposes a method for generating time-series data based on GANs a…
Baylor College of Medicine（贝勒医学院）；Rice University（莱斯大学）
A major obstacle to developing artificial intelligence applications capable of true lifelong learning is that artificial neural networks quickly or catastrophically forget previously learned tasks when trained on a new one. Numerous methods for alleviating catastrophic forgetting are currently being proposed, but differences in evaluation protocols make it difficult to directly compare their performance. To enable more meaningful comparisons, here we identified three distinct scenarios for continual learning based on whether task identity is known and, if it is not, whether it needs to be infe…
Stanford University（斯坦福大学）；University of Mumbai（孟买大学）
Recently, learning-based models have enhanced the performance of single-image super-resolution (SISR). However, applying SISR successively to each video frame leads to a lack of temporal coherency. Convolutional neural networks (CNNs) outperform traditional approaches in terms of image quality metrics such as peak signal to noise ratio (PSNR) and structural similarity (SSIM). However, generative adversarial networks (GANs) offer a competitive advantage by being able to mitigate the issue of a lack of finer texture details, usually seen with CNNs when super-resolving at large upscaling factors.…
Recovering images from undersampled linear measurements typically leads to an ill-posed linear inverse problem, that asks for proper statistical priors. Building effective priors is however challenged by the low train and test overhead dictated by real-time tasks; and the need for retrieving visually "plausible" and physically "feasible" images with minimal hallucination. To cope with these challenges, we design a cascaded network architecture that unrolls the proximal gradient iterations by permeating benefits from generative residual networks (ResNet) to modeling the proximal operator. A mix…
Graph generation with Machine Learning is an open problem with applications in various research fields. In this work, we propose to cast the generative process of a graph into a sequential one, relying on a node ordering procedure. We use this sequential process to design a novel generative model composed of two recurrent neural networks that learn to predict the edges of graphs: the first network generates one endpoint of each edge, while the second network generates the other endpoint conditioned on the state of the first. We test our approach extensively on five different datasets, comparin…
Memorial University of Newfoundland（纽芬兰纪念大学）；University of Waterloo（滑铁卢大学）
In this paper we present a method for algorithmic melody generation using a generative adversarial network without recurrent components. Music generation has been successfully done using recurrent neural networks, where the model learns sequence information that can help create authentic sounding melodies. Here, we use DC-GAN architecture with dilated convolutions and towers to capture sequential information as spatial image information, and learn long-range dependencies in fixed-length melody forms such as Irish traditional reel.
The capability of recurrent neural networks to approximate trajectories of a random dynamical system, with random inputs, on non-compact domains, and over an indefinite or infinite time horizon is considered. The main result states that certain random trajectories over an infinite time horizon may be approximated to any desired accuracy, uniformly in time, by a certain class of deep recurrent neural networks, with simple feedback structures. The formulation here contrasts with related literature on this topic, much of which is restricted to compact state spaces and finite time intervals. The m…
LyapuFlow: Controlling Generative Flows with Lyapunov Feedback for Inverse Problems
Lawrence Berkeley National Lab（劳伦斯伯克利国家实验室）；Georgia Institute of Technology（佐治亚理工学院）；Siemens（西门子公司）；International Computer Science Institute（国际计算机科学研究所）
Faculty of Physics and Astronomy, Heidelberg University（海德堡大学物理与天文学学院）；Institute of Neuroinformatics, University of Zurich and ETH Zurich（苏黎世大学与苏黎世联邦理工学院神经信息学研究所）
A major tenet in theoretical neuroscience is that cognitive and behavioral processes are ultimately implemented in terms of the neural system dynamics. Accordingly, a major aim for the analysis of neurophysiological measurements should lie in the identification of the computational dynamics underlying task processing. Here we advance a state space model (SSM) based on generative piecewise-linear recurrent neural networks (PLRNN) to assess dynamics from neuroimaging data. In contrast to many other nonlinear time series models which have been proposed for reconstructing latent dynamics, our mode…
A Markovian View of Iterative-Feedback Loops in Image Generative Models: Neural Resonance and Model Collapse
Luddy School of Informatics, Computing, and Engineering, Indiana University, Bloomington, IN, USA（印第安纳大学信息学、计算与工程学院）；Department of Computer Science, Stonehill College, Easton, MA, USA（石溪学院计算机科学系）
Xidian University（西安电子科技大学）；The University of Texas at Austin（德克萨斯大学奥斯汀分校）
We develop a recurrent gamma belief network (rGBN) for radar automatic target recognition (RATR) based on high-resolution range profile (HRRP), which characterizes the temporal dependence across the range cells of HRRP. The proposed rGBN adopts a hierarchy of gamma distributions to build its temporal deep generative model. For scalable training and fast out-of-sample prediction, we propose the hybrid of a stochastic-gradient Markov chain Monte Carlo (MCMC) and a recurrent variational inference model to perform posterior inference. To utilize the label information to extract more discriminative…
Access to high-quality education at scale is limited by the difficulty of providing student feedback on open-ended assignments in structured domains like computer programming, graphics, and short response questions. This problem has proven to be exceptionally difficult: for humans, it requires large amounts of manual work, and for computers, until recently, achieving anything near human-level accuracy has been unattainable. In this paper, we present generative grading: a novel computational approach for providing feedback at scale that is capable of accurately grading student work and providin…
Peking University（北京大学）；Singapore University of Technology and Design（新加坡科技设计大学）
For health prognostic task, ever-increasing efforts have been focused on machine learning-based methods, which are capable of yielding accurate remaining useful life (RUL) estimation for industrial equipment or components without exploring the degradation mechanism. A prerequisite ensuring the success of these methods depends on a wealth of run-to-failure data, however, run-to-failure data may be insufficient in practice. That is, conducting a substantial amount of destructive experiments not only is high costs, but also may cause catastrophic consequences. Out of this consideration, an enhanc…
Multivariate time-series data are used in many classification and regression predictive tasks, and recurrent models have been widely used for such tasks. Most common recurrent models assume that time-series data elements are of equal length and the ordered observations are recorded at regular intervals. However, real-world time-series data have neither a similar length nor a same number of observations. They also have missing entries, which hinders the performance of predictive tasks. In this paper, we approach these issues by presenting a model for the combined task of imputing and predicting…
Université de Montréal（蒙特利尔大学）；Montreal Institute for Learning Algorithms（蒙特利尔学习算法研究所）；University of Bern（伯尔尼大学）；Canadian Institute for Advanced Research（加拿大高等研究院）
We consider deep multi-layered generative models such as Boltzmann machines or Hopfield nets in which computation (which implements inference) is both recurrent and stochastic, but where the recurrence is not to model sequential structure, only to perform computation. We find conditions under which a simple feedforward computation is a very good initialization for inference, after the input units are clamped to observed values. It means that after the feedforward initialization, the recurrent network is very close to a fixed point of the network dynamics, where the energy gradient is 0. The ma…
University of Toronto（多伦多大学）；Uber ATG Toronto（Uber多伦多先进技术集团）；Vector Institute（向量研究所）；DeepMind；Stanford University（斯坦福大学）；McGill University（麦吉尔大学）；Mila – Quebec Artificial Intelligence Institute（Mila - 魁北克人工智能研究所）；Canadian Institute for Advanced Research（加拿大高等研究院）
We propose a new family of efficient and expressive deep generative models of graphs, called Graph Recurrent Attention Networks (GRANs). Our model generates graphs one block of nodes and associated edges at a time. The block size and sampling stride allow us to trade off sample quality for efficiency. Compared to previous RNN-based graph generative models, our framework better captures the auto-regressive conditioning between the already-generated and to-be-generated parts of the graph using Graph Neural Networks (GNNs) with attention. This not only reduces the dependency on node ordering but …
Massachusetts Institute of Technology（麻省理工学院）；Google Brain（谷歌大脑）
In the quest towards general artificial intelligence (AI), researchers have explored developing loss functions that act as intrinsic motivators in the absence of external rewards. This paper argues that such research has overlooked an important and useful intrinsic motivator: social interaction. We posit that making an AI agent aware of implicit social feedback from humans can allow for faster learning of more generalizable and useful representations, and could potentially impact AI safety. We collect social feedback in the form of facial expression reactions to samples from Sketch RNN, an LST…
We observe that the standard log likelihood training objective for a Recurrent Neural Network (RNN) model of time series data is equivalent to a variational Bayesian training objective, given the proper choice of generative and inference models. This perspective may motivate extensions to both RNNs and variational Bayesian models. We propose one such extension, where multiple particles are used for the hidden state of an RNN, allowing a natural representation of uncertainty or multimodality.
Google Brain（谷歌大脑）；NNAISENSE；The Swiss AI Lab, IDSIA (USI & SUPSI)（瑞士人工智能实验室IDSIA（提契诺大学与瑞士南方应用科学大学））
A generative recurrent neural network is quickly trained in an unsupervised manner to model popular reinforcement learning environments through compressed spatio-temporal representations. The world model's extracted features are fed into compact and simple policies trained by evolution, achieving state of the art results in various environments. We also train our agent entirely inside of an environment generated by its own internal world model, and transfer this policy back into the actual environment. Interactive version of paper at https://worldmodels.github.io
