Emergence of grid-like representations by training recurrent neural networks to perform spatial localization

Cueva, Christopher J.; Wei, Xue-Xin

doi:10.48550/arxiv.1803.07770

Cited by 34 publications

(45 citation statements)

References 60 publications

(60 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Future work combining different motion estimation modalities such as linear/angular velocities with visual representations is likely to be considered. However, this could potentially increase the network complexity and training requirements [80], [89], especially when using real data. Quantifying the relationship between required RL performance, visual place recognition generalization capabilities, and motion estimation quality can also provide new insights for selecting between different motion estimation sensor modalities for a specific robotic navigation system.…”

Section: Discussionmentioning

confidence: 99%

MVP: Unified Motion and Visual Self-Supervised Learning for Large-Scale Robotic Navigation

Chancán,

Milford

2020

Preprint

View full text Add to dashboard Cite

Autonomous navigation emerges from both motion and local visual perception in real-world environments. However, most successful robotic motion estimation methods (e.g. VO, SLAM, SfM) and vision systems (e.g. CNN, visual place recognition-VPR) are often separately used for mapping and localization tasks. Conversely, recent reinforcement learning (RL) based methods for visual navigation rely on the quality of GPS data reception, which may not be reliable when directly using it as ground truth across multiple, month-spaced traversals in large environments. In this paper, we propose a novel motion and visual perception approach, dubbed MVP, that unifies these two sensor modalities for large-scale, target-driven navigation tasks. Our MVP-based method can learn faster, and is more accurate and robust to both extreme environmental changes and poor GPS data than corresponding vision-only navigation methods. MVP temporally incorporates compact image representations, obtained using VPR, with optimized motion estimation data, including but not limited to those from VO or optimized radar odometry (RO), to efficiently learn self-supervised navigation policies via RL. We evaluate our method on two large real-world datasets, Oxford Robotcar and Nordland Railway, over a range of weather (e.g. overcast, night, snow, sun, rain, clouds) and seasonal (e.g. winter, spring, fall, summer) conditions using the new CityLearn framework; an interactive environment for efficiently training navigation agents. Our experimental results, on traversals of the Oxford RobotCar dataset with no GPS data, show that MVP can achieve 53% and 93% navigation success rate using VO and RO, respectively, compared to 7% for a vision-only method. We additionally report a trade-off between the RL success rate and the motion estimation precision, suggesting that visiononly navigation systems can benefit from using precise motion estimation techniques to improve their overall performance.

show abstract

Section: Discussionmentioning

confidence: 99%

MVP: Unified Motion and Visual Self-Supervised Learning for Large-Scale Robotic Navigation

Chancán,

Milford

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…For these actions, each can be thought of as a velocity of sorts, with the basis vector being the fly's own body axis. Cueva and Wei (2018) found that modeling movement using velocities leads to the emergence of neurological grid cells resemblance in the RNN parametrization, which provides a rationale for this encoding.…”

Section: Spatial Localizationmentioning

confidence: 94%

Imitation Learning of Factored Multi-agent Reactive Models

Teng,

Le,

Scibior

et al. 2019

Preprint

View full text Add to dashboard Cite

We apply recent advances in deep generative modeling to the task of imitation learning from biological agents. Specifically, we apply variations of the variational recurrent neural network model to a multi-agent setting where we learn policies of individual uncoordinated agents acting based on their perceptual inputs and their hidden belief state. We learn stochastic policies for these agents directly from observational data, without constructing a reward function. An inference network learned jointly with the policy allows for efficient inference over the agent's belief state given a sequence of its current perceptual inputs and the prior actions it performed, which lets us extrapolate observed sequences of behavior into the future while maintaining uncertainty estimates over future trajectories. We test our approach on a dataset of flies interacting in a 2D environment, where we demonstrate better predictive performance than existing approaches which learn deterministic policies with recurrent neural networks. We further show that the uncertainty estimates over future trajectories we obtain are well calibrated, which makes them useful for a variety of downstream processing tasks.

show abstract

“…Recurrent neural networks are common machine learning tools to process sequences, such as speech and text. In neuroscience, they have been used to model various aspects of the cognitive and motor systems [Mante et al, 2013, Sussillo et al, 2015, Cueva and Wei, 2018. Unlike convolutional networks used to model visual systems that are trained on large-scale image classification tasks, recurrent networks are usually trained on the specific cognitive or motor tasks that neuroscientists are studying.…”

Section: Recurrent Neural Network For Cognitive and Motor Systemsmentioning

confidence: 99%

Artificial neural networks for neuroscientists: A primer

Yang,

Wang

2020

Preprint

View full text Add to dashboard Cite

Artificial neural networks (ANNs) are essential tools in machine learning that are increasingly used for building computational models in neuroscience. Besides being powerful techniques for data analysis, ANNs provide a new approach for neuroscientists to build models that capture complex behaviors, neural activity and connectivity, as well as to explore optimization in neural systems. In this pedagogical Primer, we introduce conventional ANNs and demonstrate how they have been deployed to study neuroscience questions. Next, we detail how to customize the analysis, structure, and learning of ANNs to better address a wide range of challenges in brain research. To help the readers garner hands-on experience, this Primer is accompanied with tutorial-style code in PyTorch and Jupyter Notebook, covering major topics. Artificial neural networks in neuroscienceArtificial neural networks (ANNs), or deep neural networks, have emerged as a dominant framework in machine learning (ML) nowadays [LeCun et al., 2015], leading to breakthroughs across a wide range of applications, including computer vision [Krizhevsky et al., 2012], natural language processing [Devlin et al., 2018], and strategic games [Silver et al., 2017]. Some key ideas in this field can be traced to brain research: supervised learning rules have its root in the theory of training perceptrons which in turn was inspired by the brain [Rosenblatt, 1962]; the hierarchical architecture [Fukushima and Miyake, 1982] and convolutional principle [LeCun and Bengio, 1995] were closely linked to our knowledge about the primate visual system Wiesel, 1962, Felleman andVan Essen, 1991]. Today, there is a continued exchange of ideas from neuroscience to the field artificial intelligence [Hassabis et al., 2017].At the same time, machine learning offers new and powerful tools for systems neuroscience. One utility of the deep learning framework is to analyze neuroscientific data (Figure 1A). In particular, the advances in computer vision makes convolutional neural networks useful for processing a range of image and video data. Behaviors over time, for instance micro-movements of animals in a laboratory experiment, are Preprint.

show abstract

Emergence of grid-like representations by training recurrent neural networks to perform spatial localization

Cited by 34 publications

References 60 publications

MVP: Unified Motion and Visual Self-Supervised Learning for Large-Scale Robotic Navigation

MVP: Unified Motion and Visual Self-Supervised Learning for Large-Scale Robotic Navigation

Imitation Learning of Factored Multi-agent Reactive Models

Artificial neural networks for neuroscientists: A primer

Contact Info

Product

Resources

About