The Grid 2 Sensory Software Crack
Arnau Cyr
The representation of an animal's position in the medial entorhinal cortex (MEC) is distributed across several modules of grid cells, each characterized by a distinct spatial scale. The population activity within each module is tightly coordinated and preserved across environments and behavioral states. Little is known, however, about the coordination of activity patterns across modules. We analyzed the joint activity patterns of hundreds of grid cells simultaneously recorded in animals that were foraging either in the light, when sensory cues could stabilize the representation, or in darkness, when such stabilization was disrupted. We found that the states of different modules are tightly coordinated, even in darkness, when the internal representation of position within the MEC deviates substantially from the true position of the animal. These findings suggest that internal brain mechanisms dynamically coordinate the representation of position in different modules, ensuring that they jointly encode a coherent and smooth trajectory.
Microelectrocorticography (µECoG) provides insights into the cortical organization with high temporal and spatial resolution desirable for better understanding of neural information processing. Here we evaluated the use of µECoG for detailed cortical recording of somatosensory evoked potentials (SEPs) in an ovine model. The approach to the cortex was planned using an MRI-based 3D model of the sheep's brain. We describe a minimally extended surgical procedure allowing placement of two different µECoG grids on the somatosensory cortex. With this small craniotomy, the frontal sinus was kept intact, thus keeping the surgical site sterile and making this approach suitable for chronic implantations. We evaluated the procedure for chronic implantation of an encapsulated µECoG recording system. During acute and chronic recordings, significant SEP responses in the triangle between the ansate, diagonal, and coronal sulcus were identified in all animals. Stimulation of the nose, upper lip, lower lip, and chin caused a somatotopic lateral-to-medial, ipsilateral response pattern. With repetitive recordings of SEPs, this somatotopic pattern was reliably recorded for up to 16 weeks. The findings of this study confirm the previously postulated ipsilateral, somatotopic organization of the sheep's sensory cortex. High gamma band activity was spatially most specific in the comparison of different frequency components of the somatosensory evoked response. This study provides a basis for further acute and chronic investigations of the sheep's sensory cortex by characterizing its exact position, its functional properties, and the surgical approach with respect to macroanatomical landmarks.
Palo Alto Research Center (PARC) and its research partners will develop, prototype, and demonstrate TRANSENSOR, an innovative, low-cost optically-based monitoring system that will increase the capacity of grid infrastructure to accommodate accelerating the integration of DER.
The University of Texas at Austin and its research partners will leverage existing and emerging sensor measurements to enhance data-driven observability and develop robust estimation and identification techniques to enable real-time grid-wise monitoring and modeling of loads and distributed energy resources.
After the discovery of grid cells, which are an essential component to understand how the mammalian brain encodes spatial information, three main classes of computational models were proposed in order to explain their working principles. Amongst them, the one based on continuous attractor networks (CAN), is promising in terms of biological plausibility and suitable for robotic applications. However, in its current formulation, it is unable to reproduce important electrophysiological findings and cannot be used to perform path integration for long periods of time. In fact, in absence of an appropriate resetting mechanism, the accumulation of errors over time due to the noise intrinsic in velocity estimation and neural computation prevents CAN models to reproduce stable spatial grid patterns. In this paper, we propose an extension of the CAN model using Hebbian plasticity to anchor grid cell activity to environmental landmarks. To validate our approach we used as input to the neural simulations both artificial data and real data recorded from a robotic setup. The additional neural mechanism can not only anchor grid patterns to external sensory cues but also recall grid patterns generated in previously explored environments. These results might be instrumental for next generation bio-inspired robotic navigation algorithms that take advantage of neural computation in order to cope with complex and dynamic environments.
Figure 1. Neural simulation modules. The Grid Network receives robot velocity information from a tracking system. In addition it receives sensory information from the Sensory Map by means of excitatory plastic projections. The activation of the Sensory Map units depends on visual information provided by a camera on board the mobile robot.
Figure 2. Encoding of sensory information. (A) Example of sensory map activation when the robot is moving toward marker #9. (B) Spatial activation (in red) for the sensory unit that encodes the distance between the robot and marker #9 approximately equal to 0.45 m.
Figure 3. Simulation of robotic experiments. The trajectory of the robot is randomly generated within the boundaries of a circular arena. A virtual camera on top of the robot records the relative position within its field of view of virtual markers arranged in a rectangular grid.
Figure 5. Components of the robotic setup. (A) Mobile robotic platform used to record sensory data during the exploration of a circular arena. A wireless camera pointing to the ceiling is mounted on top of the robot. Three reflective markers fixed on the robot frame are used by the tracking system to estimate the velocity and orientation of the robot. (B) The field of view of the camera mounted on top of the robot. The center of the camera's field of view is aligned with the center of the robot. The visible markers, arranged on a regular grid on the ceiling, are detected and decoded by the software Aruco (Garrido-Jurado et al., 2014).
Figure 6. The activation of the stabilization mechanism prevents the drift of spatial grid patterns. (A) Spiking activity of a simulated grid cell without stabilization mechanism while a simulated robot explores a circular arena for 120 s (spiking activity in red, simulated trajectory in gray). (B) Spiking activity of the same cell of panel (A) after 10 min (spiking activity in blue). (C) The activities shown in panels (A,B) Do not overlap due to the accumulation of path integration errors. (D) Spiking activity of a simulated grid cell while a simulated robot explores a circular arena for 120 s with the activation of the stabilization mechanism. (E) Spiking activity of the same cell of panel (D) after 10 min. (F) The activities shown in panels (D,E) Overlap due to the Hebbian plasticity-based stabilization mechanism.
Figure 7. Comparison of grid cell activity with and without stabilization mechanism for grid cell simulations with simulated data in input. The spiking activities (red dots) of representative simulated grid cells are shown in correspondence to the position of a virtual robot (gray trace) while exploring for 30 min a circular arena without (A) and with (B) realignment mechanism (gridness scores: -0.063 for A, 0.797 for B). The activation of this plasticity-based stabilization mechanism successfully anchors the neural activity of grid cells to external sensory cues. This results in a well-defined grid pattern in space over long periods of neural simulation.
Figure 8. Excitatory sensory currents push the activity of grid network toward the right configuration. (A) Example of excitatory currents due to learned projections from the sensory map units to the grid cells after 30 min of robotic exploration of the arena: the bumps of excitation are arranged in a grid similar to that of the grid network activity (shown in B) but are not so well-defined. (B) Example of firing rate map for a network of grid cells. The spacing of the bumps in the grid network activity is related but not equal to the spacing of the grid pattern generated in space.
Figure 9. Gridness score average with and without stabilization mechanism for grid cell simulations with simulated (left) and real data (right) in input. For each experimental session we considered the activity of one single grid cell in the middle of the network. In both simulated and real conditions there is a significant difference in the average gridness score [one-way ANOVA, F(3, 396) = 234.8 with Bonferroni correction, **p < 0.01].
Figure 11. Gridness score average as a function of (A) sensory current gain and (B) average number of visible markers. The shadow areas represent the standard error of the mean. (A) The stabilization mechanism works best if velocity-dependent currents and sensory currents are balanced. (B) The optimal current sensory gain k depends on the average number of visible markers. The stabilization performance gets worse when the current sensory gain is constant (blue line) than when it is inversely proportional to the average number of visible markers (red line).
Figure 12. Distribution of grid orientations and normalized spatial phases for spatial grids (with gridness score >0) generated by grid cells at the center of each simulated network. Distributions refer to both simulated (top panels) and real data (bottom panels) in input, initializing the connectivity from the sensory map to the grid network to a zero matrix (in blue) and to a previously learned connectivity (in red). In both simulated and real conditions the previously learned connectivity drastically reduced the possible grid orientations and phases.
aa06259810