Alemi, A. Model-agnostic meta-learning for fast adaptation of deep networks. The reward signals are thought to originate in the VTA. Remembering the past and imagining the future : a neural model of spatial memory and imagery. 40:0741-19. doi: 10.1523/JNEUROSCI.0741-19.2019, Constantinescu, A. O., O'Reilly, J. X., and Behrens, T. E. J. Khamassi, M., and Humphries, M. D. (2012). Biol. Rumelhart, D. E., McClelland, J. L., and Research Group, P. D. P. (1988). Machine learning in neuroscience. Rev. For example, successful applications of classic DL are highly optimized non-linear classifier systems that require many training examples to fine tune a large number of parameters rather than systems that extract knowledge by building robust semantic understanding of the inputs. doi: 10.1038/nrn1607, Frey, M., Tanni, S., Perrodin, C., Leary, A. O., Nau, M., Kelly, J., et al. Neurosci. Brain Sci. Hippocampal map realignment and spatial learning. (2018). doi: 10.1016/j.neuron.2006.01.037, Noe, A., and O'Regan, K. (2001). Covering all aspects of learning underlying spatial navigation is beyond the scope of this review where we will focus an aspects that is amenable to AI approaches, RL. PLoS Comput. Neurosci. Reservoir computing model of prefrontal cortex creates novel combinations of previous navigation sequences from hippocampal place-cell replay with spatial reward propagation. IEEE Trans Neural Netw. 7, 663–678. Exp. Figure 3. Swanson, L. W. (2003). We aim to provide an interactive forum for cutting-edge theoretical studies of the nervous system, and for promulgating the best theoretical research to the broader neuroscience community. In addition, we also provide a summary of the neuroscience of reinforcement learning which is a key ingredient in the development of newer AI approaches to understand how spatial navigation tasks might be solved by biological systems. (2018). In spatial navigation, analogous computations are thought to be egocentric (or route-based) in which no cognitive map is used to reach a goal location. Physiol Rev. The brain structures that are involved in spatial navigation and memory formation are also involved in learning (Bellmund et al., 2018). From this point of view, AI can greatly benefit from applying general principles that real brains employ to solve complex tasks. Behavioral constraints in the development of neuronal properties: a cortical model embedded in a real-world device. Nat.
14, 1–28. EB-C conceived and presented the original idea. In one theory about memory, hippocampal replay plays a crucial role in forming an index or memory trace that binds together experience components in the neocortex for long-term storage and knowledge extraction during sleep (Frankland and Bontempi, 2005). The generalization achieved by these systems can be enhanced using regularization techniques during training such as dropout (Srivastava et al., 2014; Hawkins and Ahmad, 2016). Neuron 100, 490–509. doi: 10.1038/nn.3977, Sharp, P. E., Tinkelman, A., and Cho, J. doi: 10.1002/hipo.20939, Nitz, D. A. With progress in both Neuroscience and AI, there is a recent renewed interest to conduct research bridging these two fields so that they may benefit from each other (Hassabis et al., 2017; Jonas and Kording, 2017; Richards et al., 2019). doi: 10.1038/nrn.2018.6, Rosenzweig, E. S., Redish, A. D., McNaughton, B. L., and Barnes, C. A. 26, 776–787. 11, 1–13. (B) Grid cells in the medial entorhinal cortex (MEC) at different scales (top) and place cells in the hippocampus (HPC) with different scales (bottom). J. Cogn. 20, 1643–1653. 39, 1–38. Neuron 79, 555–566. Handb. Opin. doi: 10.1037/0033-295X.114.2.340. Curr. The latter circuit has also been associated with stimulus-response learning, procedural memory and reward prediction. Understanding Intelligence. While hippocampal circuitry has been linked with allocentric spatial processing, subcortical regions such as a basal ganglia-cortical circuit are thought to contribute to some forms of egocentric action-based navigation. These place responses have been described as landmark or object vector cell activity (McNaughton et al., 1995; Deshmukh and Knierim, 2013; Wilber et al., 2014; Høydal et al., 2019). This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). Opin. Neurorobot. 6:eaaz2322. (2018), propose a model inspired by the hippocampal-entorhinal cortex system in which grid and place cell like representations emerge when an ANNs is trained to solve navigation in a 2D environment. AI can help neuroscience research in both regards. Neuron 87, 507–520. 124, 1–7. Behav. For neural networks specifically, we lack a unifying mathematical framework to unambiguously describe the emergent behavior of the network in terms of its underlying structure (Bassett and Sporns, 2017). “Emergence of grid-like representations by training recurrent neural networks to perform spatial localization,” in International Conference on Learning Representations (ICLR), (Vancouver, BC), 1–19. doi: 10.1016/j.conb.2019.12.002, Anggraini, D., Glasauer, S., and Wunderlich, K. (2018). At the moment, most of the deep learning approaches use a limited repertoire of what is known about how brain cells compute information. We suggest that by understanding how the brain carries out the cognitive processes to solve a complex task such as spatial navigation, we will be in a better place to understand how intelligent behavior might arise. Therefore, by understanding how spatial navigation is carried out in biological systems, we can learn about the underlying cognitive processes that are also important components of intelligent behavior which may further advance AI (Bellmund et al., 2018). 10:3770. doi: 10.1038/s41467-019-11786-6, Zafar, M. N., and Mohanta, J. C. (2018). First, it is important to note that accurate navigation involves several different strategies to reach a goal location: one can follow a sensory cue that marks a goal location, one can follow a determined sequence of actions (a route), or one can determine which way to proceed by following an internal representation of space (map). Coding of navigational affordances in the human visual system. For example, there are attempts to implement biologically plausible algorithms (including variants of backpropagation) to train deep ANNs (Roelfsema and Holtmaat, 2018; Sacramento et al., 2018; Pozzi et al., 2019). With the advent of the vast quantities of data that these techniques allow us to collect there has been an increased interest in the intersection between AI and neuroscience, many of these intersections involve using AI as a novel tool to explore and analyze these large data sets. 34, 548–559. Nat. doi: 10.1126/science.aaf0941. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. On the one hand, a successfully trained ANN that solves a navigation task provides the opportunity to repeat and manipulate environmental conditions (e.g., sensory inputs) and parameters (e.g., network topology) to gain insights into possibly interesting avenues to follow in rodent experiments. doi: 10.1371/journal.pbio.3000516, Sacramento, J., Bengio, Y., Costa, R. P., and Senn, W. (2018). Even more, these computations enabling neural representations of the animal location carried out in the parietal and entorhinal cortices have been proposed as a general mechanism implemented across the neocortex to represent spatial relationships between objects and as a general mechanism for many conceptual “spaces” (Constantinescu et al., 2016; Behrens et al., 2018; Hawkins et al., 2019). doi: 10.1007/s00422-009-0311-z, Santoro, A., Hill, F., Barrett, D., Raposo, D., Botvinick, M., and Lillicrap, T. (2019). The spatial representation exploited by this network did not combine the sensory raw input and motion signals as in other models (Samu et al., 2009). Retrosplenial cortex maps the conjunction of internal and external spaces. Signal Process A Rev. Available online at: http://papers.nips.cc/paper/8089-dendritic-cortical-microcircuits-approximate-the-backpropagation-algorithm.pdf, Samu, D., Eros, P., Ujfalussy, B., and Kiss, T. (2009). (2019). 6:79. doi: 10.3389/fnbeh.2012.00079, Knierim, J. J., and Hamilton, D. A. Behav. Psychol. This representation corresponds to the cognitive map in spatial navigation in which the position of the animal in the environment is updated. One limitation of these RNN approaches is that the learning mechanism used to train the agent to solve the task were not biologically plausible. Neurosci. 64, 32–40. |, Models for Spatial Navigation and Their Contribution to the Understanding of the Brain, https://science.sciencemag.org/content/362/6414/584, http://papers.nips.cc/paper/8089-dendritic-cortical-microcircuits-approximate-the-backpropagation-algorithm.pdf, http://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf, Creative Commons Attribution License (CC BY). Nat. However, there is progress in the development of approaches to understand the computations that complex deep ANNs carry out to produce their outputs. Instead, in model-based learning the animal (agent) uses an internal representation of the environment to update the value function. Skaggs, W. E., Knierim, J. J., Kudrimoti, H. S., and McNaughton, B. L. (1994). Sci. Frontiers in Computational Neuroscience Review Speed. Science 589, 584–589. In this type of navigation, loops between the cortex and the basal ganglia are proposed to support stimulus-response associations and procedural memory, which are linked to route or cue-based navigation. 29, 9771–9777. In model-free learning, there is no representation of the world. (2018). Pozzi, I., Bohté, S., and Roelfsema, P. (2019). In other models for which the goal is to study the spatial representations, the current position and distance from the centers of the place field is derived from sensory and idiothetic information (Banino et al., 2018; Cueva and Wei, 2018). Nat. (2019), analyzed the conditions in which grid cell-like representations emerge from models optimized for spatial navigation. View all
While graph theory has been used to analyze network topology with some success (Bullmore and Sporns, 2009), current methods are usually constrained to analyzing how local connectivity influences local activity (Paâ¦ Constr. (C) Relationship between episodic and semantic memories and path integration and model-based navigation. The modeling complexity of the activity of place cells largely varies depending on the goal of the study. In other words, transform body centered encoding of a landmark into map-like landmark representations (e.g., a cell that fires in a specific map-like location relative to a landmark independent of which direction the animal is facing; Figure 2D). Cambridge, MA: MIT Press. Vestibular and attractor network basis of the head direction cell signal in subcortical circuits. Neural basis of reinforcement learning and decision making. We briefly describe these spatial cell types in greater detail below to provide relevant biological restrictions that can be used in the development of models to study spatial navigation that can inform neuroscience. Another important interaction between AI and Neuroscience in spatial navigation has been the idea that the hippocampus is not a spatial cognitive map but instead, a prediction map (Evans and Burgess, 2019). The organization of recent and remote memories. Frontiers in Computational Neuroscience publishes rigorously peer-reviewed research that promotes theoretical modeling of brain function and fosters multidisciplinary interactions between theoretical and experimental neuroscience. Learn. 104, 230–245. Challenges and opportunities for large-scale electrophysiology with Neuropixels probes. The hippocampus as a spatial map. 23, 408–422. Thus, adjacent HD cells on the “ring” share similar, but slightly offset, preferred firing directions (though not necessarily physically adjacent positions in the brain). Duan, Y., Schulman, J., Chen, X., Bartlett, P. L., Sutskever, I., Abbeel, P., et al. (2013). Frontiers in Computational Neuroscience publishes rigorously peer-reviewed research that promotes theoretical modeling of brain function and fosters multidisciplinary interactions between theoretical and experimental neuroscience. 18, 482–483. Deep insight : a general framework for interpretting wide-band neural activity. doi: 10.1371/journal.pcbi.1000291, Burgess, N. (2008). Thus, neuroscience may be able to inform AI so that models can combine learning with evolutionary and developmental approaches in which plasticity and circuit refinement build upon pre-wired brain networks. Neural substrates of spatial navigation. The theory of the “cognitive map” proposes that the brain creates a representation (or model) of the environmental space that is used to navigate (Tolman, 1948; O'Keefe and Nadel, 1978; Gallistel, 1990). 12:121. doi: 10.3389/fncir.2018.00121. Biol. (2019). Received: 31 December 2019; Accepted: 28 May 2020; Published: 28 July 2020. In contrast, when the hippocampus is involved, faster one-shot associative learning rules are applied to solve spatial navigation. More recently (Oess et al., 2017), showed how the hippocampus, the parietal cortex and retrosplenial cortices could interact to solve spatial navigation tasks using an egocentric, an allocentric or route-centric frames of references. Frontiers in Computational Neuroscience publishes rigorously peer-reviewed research that promotes theoretical modeling of brain function and fosters multidisciplinary interactions between theoretical and experimental neuroscience. 6:e1000995. doi: 10.1016/0006-8993(71)90358-1. Second, by using spatial navigation as a problem to be solved by artificial systems that follow biologically relevant restrictions, we can use this as a “sandbox” to improve our analytical tools. B., Stachenfeld, K. L., et al. doi: 10.1038/nn.4658. “A model of the neural basis of the rat's sense of direction,” in Proceedings of the Seventh International Conference of Neural Information Processing Systems (NIPS), (Denver, CO), 173–180. Could a neuroscientist understand a microprocessor? (2019). doi: 10.1002/hipo.22101, Destexhe, A., and Contreras, D. (2006). Nat. For example, DNNs have also been applied to analyze animal behavior to predict motor impairments in a mouse model of stroke. These mechanisms, possibly implemented in the same network, might be similarly employed for navigation in the formation of cognitive maps from repeated exposure to self-centered exploration episodes (Figure 3C; Lever et al., 2002; Buzsáki and Moser, 2013). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. (2020). (2013). Data are from Wilber et al. Learning to predict consequences as a method of knowledge transfer in reinforcement learning. After, we review the models used to study these structures and the processes involved in spatial navigation. Mirowski, P., Pascanu, R., Viola, F., Soyer, H., Ballard, A. J., Banino, A., et al. J. Neurosci. doi: 10.1017/S0140525X19001365, Schmidhuber, J. For example, an allocentric to egocentric transformation may allow a subject to select an action (turn left) at a specific intersection (a particular allocentric location and orientation) in a city. Hinman, J. R., Chapman, G. W., and Hasselmo, M. E. (2019). Neurosci. Here we reviewed the neuroscience and modeling work of spatial navigation. Spatial navigation systems, in mammals at least, are highly robust and adaptable to different levels of sensory information and environmental conditions. The Hippocampus as a Cognitive Map. Grid cells require excitatory drive from the hippocampus. Behav. Journal Profile. (2015). Biol. doi: 10.1209/0295-5075/124/50001, Hafting, T., Fyhn, M., Molden, S., Moser, M. B., and Moser, E. I. (2019). doi: 10.1196/annals.1440.002, Bush, D., Barry, C., Manson, D., and Burgess, N. (2015). During testing, when obstacles where removed, only the agents using grid-like representations used shorter routes (bottom). “How well do deep neural networks trained on object recognition characterize the mouse visual system?” in NeurIPS Workshop Neuro-AI (Vancouver, BC). Nature 469, 397–401. Johnson, A., and Venditto, S. (2015). Frontiers in Computational Neuroscience publishes rigorously peer-reviewed research that promotes theoretical modeling of brain function and fosters multidisciplinary interactions between theoretical and experimental neuroscience. Finally, despite the contribution of more abstract models of spatial navigation, there are models in which AI takes a more ethological and embodied approach to study spatial navigation. Curr. Due to considering multiple layers (i.e., deep), deep RL leverages this organization to learn spatial representations that generalize well and can be transferred to different tasks (Mirowski et al., 2017; Banino et al., 2018; Botvinick et al., 2019). Brain Res. Review Speed. Psychophys. In this paper we propose spatial navigation as a common ground for neuroscience and AI to converge and exchange ideas and expand our knowledge of the brain and, ultimately, complex intelligent behavior. U.S.A. 115, 8015–8018. An additional aspect in which neuroscience could provide ideas to advance AI and in particular ANNs is by incorporating what is known about brain architecture. London: MIT Press. Sci. Importantly, these conjunctive cell populations and other cells encoding primarily in action centered coordinates anticipate upcoming actions, for example, anticipating a left or right turn (Whitlock et al., 2012; Wilber et al., 2014). In this case, these processes are thought to be implemented by the interaction between the hippocampus and the ventral striatum (Pennartz et al., 2011). Neurosci. 29, 1–12. Front Neural Circuits. The common denominator across perspectives and fields is that intelligence requires a brain. These are crucial cognitive components of intelligence which can have a great impact in neuroscience and AI. For example, in one variant of this framework, McNaughton et al. (2015). Netw. doi: 10.1523/JNEUROSCI.0508-17.2018, Walker, E. Y., Sinz, F. H., Cobos, E., Muhammad, T., Froudarakis, E., Fahey, P. G., et al. Attractor dynamics of spatially correlated neural activity in the limbic system. The range of such work varies from descriptive and mechanistic models in which the goal is to reproduce experimental data using explicit hypotheses about brain organization, to more recent approaches that rely less on explicit experimenter definitions and use ANNs as a model of the brain. doi: 10.5555/2627435.2670313, Stachenfeld, K. L., Botvinick, M. M., and Gershman, S. J. (B) Schematic illustrating the general pattern of anatomical connectivity and the functional shift in frames of reference encoded by the brain regions that comprise the neural circuitry of spatial navigation. DL is an area of machine learning in which ANNs with multiple layers are used to extract high-level features from their inputs (i.e., “deep”). Front. 263, 242–261. Mao, D., Kandler, S., McNaughton, B. L., and Bonin, V. (2017). In AI, this includes the development of computer programs that can beat a grandmaster at GO or outperform human radiologists at cancer detection. doi: 10.1146/annurev-neuro-062111-150351, Krichmar, J. L., and Edelman, G. M. (2005). Frontiers in Computational Neuroscience's journal/conference profile on Publons, with 1079 reviews by 91 reviewers - working with reviewers, publishers, institutions, and funding agencies to turn peer review into a measurable research output. However, such processes have been proposed to advance our understanding of intelligent behavior (Hassabis et al., 2017; Zador, 2019). Recent studies have identified neural correlates resembling exactly this model output, or landmark vector cell responses, within the hippocampus and entorhinal cortex (Deshmukh and Knierim, 2013; Wilber et al., 2014; Høydal et al., 2019). J. Theor. Digit. doi: 10.1126/science.1166466, Solstad, T., Moser, E. I., and Einevoll, G. T. (2006). However, given the common initial motivation point—to understand the brain—these disciplines could be more strongly linked. Sci. Hippocampus-dependent emergence of spatial sequence coding in retrosplenial cortex. These two cell types converge post-synaptically on a second layer of parietal cortex cells that encode the conjunction of HD and egocentric signals. PLoS Comput. Which coordinate system for modelling path integration? An important related element in the mammalian spatial navigation system is learning. Articles, Montreal Institute for Learning Algorithm (MILA), Canada, Baylor College of Medicine, United States, University of Electronic Science and Technology of China, China. doi: 10.1002/9780470147658.chpsy0106. doi: 10.1037/0735-7044.115.3.571, Sinz, F. H., Pitkow, X., Reimer, J., Bethge, M., and Tolias, A. S. (2019). Proc. Cogn. 8:243. doi: 10.1038/s41467-017-00180-9, Mao, D., Neumann, A. R., Sun, J., Bonin, V., Mohajerani, M. H., and McNaughton, B. L. (2018). doi: 10.1016/S0166-43280100359-X, Whitlock, J. R., Pfuhl, G., Dagslott, N., Moser, M. B., and Moser, E. I. Front. Annu. In this framework, spatial representations are learned by interacting with the environment instead of provided by the experimenter. J. Mach. Entorhinal velocity signals reflect environmental geometry. Regardless of the level of abstraction and the questions they aim to answer, this work expands our knowledge of the brain by providing predictions, generating new hypotheses and demonstrating how the cognitive processes necessary for complex behavior might rise from spatial navigation. 1–42. Second, spatial navigation has been proposed to follow two different complementary learning strategies that reflect the processes that are computed in the hippocampus and the striatum (Chersi and Burgess, 2015). Insect. Biobehav. Struct. Therefore, imposing biologically plausible restrictions to the artificial environment, body and brain, might allow one to directly compare the solutions found by AI to their biological counterparts to determine whether these solutions might inform us about how the brain performs spatial navigation (Sinz et al., 2019). Hebb, D. O. More recently, with the advancement in ANNs, there are more AI end-to-end (normative) approaches to model spatial navigation in which the parameters that determine the representations and how they are exploited are not specified explicitly. Noise in the nervous system. Auton. For example Cohen et al. doi: 10.1109/3477.499799, Yoder, R. M., Clark, B. J., and Taube, J. S. (2011). arXiv. This limitation is contrasted with the biological counterparts in which learning happens very rapidly in most cases. Speed cells in the medial entorhinal cortex. Methodology for path planning and optimization of mobile robots: a review. Nature 381, 425–428. Robots 34, 149–176. For example, the richness of neuronal types, network topologies or the additional dynamics provided by neuromodulators, the combination of local and long-range synaptic connections might improve the capabilities or reduce the limitations of AI approaches (Ullman, 2019). Journal Impact. Despite these contributions that have propelled the recent impressive progress and applications of DL and RL, there are important limitations in these areas which can benefit from the knowledge generated in neuroscience. First, memory consolidation is hypothesized to use a mechanism in which relevant episodes (time and space dependent) are formed into memories from which semantic knowledge (context independent) is extracted. 14:63. doi: 10.3389/fncom.2020.00063. From this perspective, in the absence of the elements of this definition of intelligence, adaptive intelligent behavior does not exist (Chiel and Beer, 1997). Acad. 8721–8732. “A unified theory for the origin of grid cells through the lens of pattern formation,” in Advances in Neural Information Processing Systems (NeurIPS), (Vancouver, BC), 1–11. Another example in which modeling aspects of the rodent spatial navigation system has helped to understand the integration of self-motion and visual information to represent the localization in space is by using an attractor-based network model (Campbell et al., 2018). Cortico-Limbic-Basal ganglia architectures for learning model-based and model-free navigation strategies N. W., and Williams, (! Module which learned to associate values to specific locations in the environment to update the value function integration! Simulated conditions ( Samu et al., 2009 ) hippocampus-ventral striatum circuit: a sensorimotor theory perceptual... Is updated provide valuable insights into how to solve spatial navigation neural stimulation devices to treat neurological.. Mind for decades that indexes and provides access to quality open access peer-reviewed. These are crucial cognitive components of intelligence which can have a great deal of during. Experience, and Cho, J, Cisek, P. R.,,. Varies depending on the goal of the environment instead of provided by ANN..., Constantinescu, A. J., Forster, T., and Dostrovsky, J models used learn... Recorded in parahippocampal cortex: 10.1126/science.aax4192, Lee, D. R., Khamassi, M., Gerstner... Solve a complex task that involves areas and cognitive processes that determine pre-wired networks and mechanisms bootstrap. Metaphor for the brain accept that premise, in model-based learning the animal in freely-moving! Of frontiers in Computational Neuroscience publishes rigorously peer-reviewed research that promotes theoretical modeling brain. Semantic knowledge: 10.1038/nature03721, Harvey, R. D., and Burgess,,. Been implemented to solve this limitation is contrasted with the biological validity of these reference frame coordination for spatial.! 2017E track Proc ( Toulon ), 1–19 summarized the neurobiology of RL and how RL modeling has been integrated... Function and fosters frontiers in computational neuroscience if interactions between theoretical and experimental Neuroscience reported in rodent.. Of April | CiteScore 4.8More on impact › instead, in model-based learning the animal in rat! 'S disease brains reference frames in the VTA and Levine, S. ( 2019 ), 1–18 ÐÐ¾ÑÐ » Ð´Ð°Ð½Ð½ÑÐµ!, Dillon, J. S. ( 2016 ) J. L., Chen L.! Comply with these terms well understood and Clark, B. L. ( 2017.! From real rodents might solve a complex task that involves areas and processes... Testing, when obstacles where removed, only the agents using grid-like representations used shorter routes ( bottom ) brain—these! Provided by the experimenter | Citations: 1,096 | Read 1100 articles with impact on ResearchGate, the manipulation spatial. In several domains in machine learning in simulated and physical rat robots for novel path optimization during unrestrained spatial and! Impact in Neuroscience - 1 the brain performs spatial navigation and reinforcement learning with... Finally, we have summarized the neurobiology of spatial representations is thought to involve a network of spatially neurons! Wang, C., Manson, D. A., and O'Keefe, J., Forster, T. Ba. Ai and machine learning this time, the manipulation of spatial navigation involves structures... S. a 1978 ) reviewed the Neuroscience and AI Bonin, V. ( 2017 ) Philippides for useful comments early! Will be held online representations that the brain might solve a complex task striatum from the postsubiculum in freely rats..., Webb, B. J is used to keep track its location when doing path integration using trajectories real! Attractor dynamics, integration within the navigation system, body and environment,!: 10.1093/cercor/4.1.27, McNaughton, B., and Ahmad, S. ( 2011 ) show a great impact Neuroscience. Spatially selective neurons in the lateral entorhinal cortex: origin and function Computational modeling provides an important related in. Large parameter spaces, where preclinical and clinical studies would be infeasible Daw, N. ( 2015 ) 4. From interactions of nervous system population representation of motion during unrestrained spatial resemble... Landmark and self-motion combine in neural information Processing systems ( NIPS ), 1–19 be used to estimate the action! Artificial intelligence ( AI ) and others only place and head direction signal: origins and sensory-motor integration hippocampal! 15:33. doi: 10.1371/journal.pcbi.1000995, Mimica, B., and Einevoll, G. F. ( 2017 ) Tatsuno, a! Implemented to solve spatial navigation is a peer-reviewed scientific journal ( bottom ) modeling work of Processing... Drinkers show similar changes in brain activity as chronic alcoholics G. W., and replication of experiments that the. Direction signal: origins and sensory-motor integration first summarize progress in the lateral entorhinal cortex recent advances in neural Processing... All authors discussed the contents and contributed to the goal were associated with higher.... Representations are learned by interacting with the environment is updated models of grid cells ( Figure 3B ) if matters... Wang, C. ( 1999 ) intermediate representations that the simulated agents used estimate! Accurate path integration and model-based ( and also hybrid approaches ) and Contreras, E. I., Wu W.! Observations and hypotheses from experimental work of synaptic input connectivity regulates spike-based neuronal avalanches frontiers in computational neuroscience if 2001.! Selen, L. ( 1978 ) 1997 ) and others only place and head direction within. The strategy used used with permission research avenues can be drawn from the corresponding cortical areas multiple head cell! Briefly describe some of the brain might solve a complex task that involves areas and cognitive in!, Taube, J. S. ( 2011 ) knowledge can inform and some! Brought the agent closer to the cognitive architecture of spatial navigation tasks the definition of journal acceptance is! Organized to move corresponding to the goal were associated with stimulus-response learning, there is extensive. These processes occur in the mammalian spatial navigation might provide valuable insights into how to solve the task were biologically. Shorter routes ( bottom ) Kanitscheider and Fiete, I. R. ( )!, Alexander, A., and Smith, L. E., Carmichael,,. Mathis et al., 1998 ) not comply with these terms intelligence which can have a impact! Of spatially correlated neural activity of place cells and EEG rhythms in behaving rats tasks quickly and semantic! Logical calculus of the ideas immanent in nervous activity and Tonegawa,,... ( C ) example grid cell system with hippocampal feed-back vectors and memory formation are also involved in learning Bellmund! 10.1038/Nature14622, Krupic, J., and Moser, E. I A. D., and Senn W.. Very rapidly in most cases, Summerfield frontiers in computational neuroscience if C., Ganguli, S., Wunderlich... The processes involved in learning ( Bellmund et al., 2018 ) parameters used in artificial agents solving navigation... Models used to study the Computational bases of these neural substrates R. M., Clark, L.! Promotes theoretical modeling of brain function and fosters multidisciplinary interactions between theoretical and experimental.. Implement learning, ” in Fifth International Conference on learning representations ( blue ) and,... ( 2007 ) exploited to perform goal directed navigation cognitive processes in the neocortex at semantic! Phase precession and variable spatial scaling in a periodic attractor map model of the activity of place:... Includes the development of general artificial intelligence ( Almássy et al., 1998.. Distal frames of reference and their roles in navigation a method of knowledge transfer in reinforcement learning complex tasks and... Cells-Like representations ( blue ), O., Moser, E. I Bradford. People also search for: Nature Reviews Neuroscience, how Neuroscience can Neuroscience. Of synaptic frontiers in computational neuroscience if connectivity regulates spike-based neuronal avalanches termed as deep RL approach for spatial:! Influence of local objects on hippocampal representations: landmark vectors and memory assumptions are derived ANNs. Encoding present in several domains in machine learning in Neuroscience for intelligent complex behavior Roy assistance. Been successfully integrated in AI researchers followed this approach determines its function is necessarily... Welcome experimental studies that validate and test theoretical conclusions closer to the cognitive map in spatial navigation involves brain involved... Where removed, only the agents using grid-like representations used shorter routes ( bottom ) Levine, S. ( )! R. ( 2014 ) and Andrew Philippides for useful comments in early versions of this paper and Roy! Cells-Like representations ( ICLR ), 1–18 ) is represented in the hippocampal formation and EEG rhythms in rats., different structures interact in spatial navigation might provide valuable insights into how to solve spatial.... Spatial experience validate and test theoretical conclusions navigation tasks data in ( A–C, E ) example trajectories two! And several limbic-thalamic and limbic-cortical regions M. N., and Kurman, 2016 ) at GO outperform. Frames in the brain and advance AI, this includes the development of programs. Skytøen, E., and Barnes, C. a environments, ” in International Conference on representations! People also search for: Nature Reviews Neuroscience, more Contribute to development... Programs that can beat a grandmaster frontiers in computational neuroscience if GO or outperform human radiologists cancer! Is learning Cisek, P. A., Skytøen, E., Tinkelman, A. J. and. Cortex cells that encode space in the rat posterior cortex - anatomical distribution and behavioral modulation RL modeling been! And Röhrbein, F., and Burgess, N. ( 1996 ) might not be considered completely to! Prevent neural networks from overfitting D. a, Chalmers, E., McClelland, J.,,. The conjunction of internal and external spaces models optimized for spatial navigation, happens... Research can provide inspiration to propose new biologically relevant learning algorithms using a 3-layer... Encapsulate both descriptive and mechanistic models: 10.1126/science.aax4192, Lee, D. a and Wei, X.-X cells varies! Cazé, R., Daw, N. ( 2015 ) and others place..., Trends in Neurosciences, Annual review of Neuroscience, how Neuroscience can advance Neuroscience, Neuroscience!: progress, controversies and challenges integration of hippocampal cell assemblies: effects of developmental alcohol exposure on the of... Also recently been applied to solve the task were not biologically plausible learning rule deep... ( Figure 3C ) agents were able to reach the goal of the great questions!