Abstract Accurate tracking of the 3D pose of animals from video recordings is critical for many behavioral studies, yet there is a dearth of publicly available datasets that the computer vision community could use for model development. We here introduce the Rodent3D dataset that records animals exploring their environment and/or interacting with each other with multiple cameras and modalities (RGB, depth, thermal infrared). Rodent3D consists of 200 min of multimodal video recordings from up to three thermal and three RGB-D synchronized cameras (approximately 4 million frames). For the task of optimizing estimates of pose sequences provided by existing pose estimation methods, we provide a baseline model called OptiPose. While deep-learned attention mechanisms have been used for pose estimation in the past, with OptiPose, we propose a different way by representing 3D poses as tokens for which deep-learned context models pay attention to both spatial and temporal keypoint patterns. Our experiments show how OptiPose is highly robust to noise and occlusion and can be used to optimize pose sequences provided by state-of-the-art models for animal pose estimation.

Oscillations are ubiquitous features of brain dynamics that undergo task-related changes in synchrony, power, and frequency. The impact of those changes on target networks is poorly understood. In this work, we used a biophysically detailed model of prefrontal cortex (PFC) to explore the effects of varying the spike rate, synchrony, and waveform of strong oscillatory inputs on the behavior of cortical networks driven by them. Interacting populations of excitatory and inhibitory neurons with strong feedback inhibition are inhibition-based network oscillators that exhibit resonance (i.e., larger responses to preferred input frequencies). We quantified network responses in terms of mean firing rates and the population frequency of network oscillation; and characterized their behavior in terms of the natural response to asynchronous input and the resonant response to oscillatory inputs. We show that strong feedback inhibition causes the PFC to generate internal (natural) oscillations in the beta/gamma frequency range (>15 Hz) and to maximize principal cell spiking in response to external oscillations at slightly higher frequencies. Importantly, we found that the fastest oscillation frequency that can be relayed by the network maximizes local inhibition and is equal to a frequency even higher than that which maximizes the firing rate of excitatory cells; we call this phenomenon population frequency resonance. This form of resonance is shown to determine the optimal driving frequency for suppressing responses to asynchronous activity. Lastly, we demonstrate that the natural and resonant frequencies can be tuned by changes in neuronal excitability, the duration of feedback inhibition, and dynamic properties of the input. Our results predict that PFC networks are tuned for generating and selectively responding to beta- and gamma-rhythmic signals due to the natural and resonant properties of inhibition-based oscillators. They also suggest strategies for optimizing transcranial stimulation and using oscillatory networks in neuromorphic engineering.

Firing fields of grid cells in medial entorhinal cortex show compression or expansion after manipulations of the location of environmental barriers. This compression or expansion could be selective for individual grid cell modules with particular properties of spatial scaling. We present a model for differences in the response of modules to barrier location that arise from different mechanisms for the influence of visual features on the computation of location that drives grid cell firing patterns. These differences could arise from differences in the position of visual features within the visual field. When location was computed from the movement of visual features on the ground plane (optic flow) in the ventral visual field, this resulted in grid cell spatial firing that was not sensitive to barrier location in modules modeled with small spacing between grid cell firing fields. In contrast, when location was computed from static visual features on walls of barriers, i.e. in the more dorsal visual field, this resulted in grid cell spatial firing that compressed or expanded based on the barrier locations in modules modeled with large spacing between grid cell firing fields. This indicates that different grid cell modules might have differential properties for computing location based on visual cues, or the spatial radius of sensitivity to visual cues might differ between modules.

This paper presents a novel place recognition algorithm inspired by the recent discovery of overlapping and multi-scale spatial maps in the rodent brain. We mimic this hierarchical framework by training arrays of Support Vector Machines to recognize places at multiple spatial scales. Place match hypotheses are then cross-validated across all spatial scales, a process which combines the spatial specificity of the finest spatial map with the consensus provided by broader mapping scales. Experiments on three real-world datasets including a large robotics benchmark demonstrate that mapping over multiple scales uniformly improves place recognition performance over a single scale approach without sacrificing localization accuracy. We present analysis that illustrates how matching over multiple scales leads to better place recognition performance and discuss several promising areas for future investigation.

Boundary vector cells in entorhinal cortex fire when a rat is in locations at a specific distance from walls of an environment. This firing may originate from memory of the barrier location combined with path integration, or the firing may depend upon the apparent visual input image stream. The modeling work presented here investigates the role of optic flow, the apparent change of patterns of light on the retina, as input for boundary vector cell firing. Analytical spherical flow is used by a template model to segment walls from the ground, to estimate self-motion and the distance and allocentric direction of walls, and to detect drop-offs. Distance estimates of walls in an empty circular or rectangular box have a mean error of less than or equal to two centimeters. Integrating these estimates into a visually driven boundary vector cell model leads to the firing patterns characteristic for boundary vector cells. This suggests that optic flow can influence the firing of boundary vector cells.

The hippocampus plays an important role in the encoding and retrieval of spatial and non-spatial memories. Much is known about the anatomical, physiological and molecular characteristics as well as the connectivity and synaptic properties of various cell types in the hippocampal circuits [1], but how these detailed properties of individual neurons give rise to the encoding and retrieval of memories remains unclear. Computational models play an instrumental role in providing clues on how these processes may take place. Here, we present three computational models of the region CA1 of the hippocampus at various levels of detail. Issues such as retrieval of memories as a function of cue loading, presentation frequency and learning paradigm, memory capacity, recall performance, and theta phase precession in the presence of dopamine neuromodulation and various types of inhibitory interneurons are addressed. The models lead to a number of experimentally testable predictions that may lead to a better understanding of the biophysical computations in the hippocampus.

Firing activity from neural ensembles in rat hippocampus has been previously used to determine an animal's position in an open environment and separately to predict future behavioral decisions. However, a unified statistical procedure to combine information about position and behavior in environments with complex topological features from ensemble hippocampal activity has yet to be described. Here we present a two-stage computational framework that uses point process filters to simultaneously estimate the animal's location and predict future behavior from ensemble neural spiking activity. First, in the encoding stage, we linearized a two-dimensional T-maze, and used spline-based generalized linear models to characterize the place-field structure of different neurons. All of these neurons displayed highly specific position-dependent firing, which frequently had several peaks at multiple locations along the maze. When the rat was at the stem of the T-maze, the firing activity of several of these neurons also varied significantly as a function of the direction it would turn at the decision point, as detected by ANOVA. Second, in the decoding stage, we developed a state-space model for the animal's movement along a T-maze and used point process filters to accurately reconstruct both the location of the animal and the probability of the next decision. The filter yielded exact full posterior densities that were highly nongaussian and often multimodal. Our computational framework provides a reliable approach for characterizing and extracting information from ensembles of neurons with spatially specific context or task-dependent firing activity.

Models of the hexagonally arrayed spatial activity pattern of grid cell firing in the literature generally fall into two main categories: continuous attractor models or oscillatory interference models. Burak and Fiete (2009, PLoS Comput Biol) recently examined noise in two continuous attractor models, but did not consider oscillatory interference models in detail. Here we analyze an oscillatory interference model to examine the effects of noise on its stability and spatial firing properties. We show analytically that the square of the drift in encoded position due to noise is proportional to time and inversely proportional to the number of oscillators. We also show there is a relatively fixed breakdown point, independent of many parameters of the model, past which noise overwhelms the spatial signal. Based on this result, we show that a pair of oscillators are expected to maintain a stable grid for approximately t = 5mu(3)/(4pisigma)(2) seconds where mu is the mean period of an oscillator in seconds and sigma(2) its variance in seconds(2). We apply this criterion to recordings of individual persistent spiking neurons in postsubiculum (dorsal presubiculum) and layers III and V of entorhinal cortex, to subthreshold membrane potential oscillation recordings in layer II stellate cells of medial entorhinal cortex and to values from the literature regarding medial septum theta bursting cells. All oscillators examined have expected stability times far below those seen in experimental recordings of grid cells, suggesting the examined biological oscillators are unfit as a substrate for current implementations of oscillatory interference models. However, oscillatory interference models can tolerate small amounts of noise, suggesting the utility of circuit level effects which might reduce oscillator variability. Further implications for grid cell models are discussed.

We simulate temporal context dependent episodic memory with resulting reversed order activity in model hippocampal regions CA3 and CA1, similar to recorded activity recently published by D.J.Foster and M.A.Wilson. Our model supposes that encoding of reversed place associations in CA3 is the cause of that activity. That proposed mechanism is the focus of this paper, and we simulate resulting activity in model layer III of entorhinal cortex, and in hippocampal regions CA3 and CA1. Learning associates spike episodes maintained in a persistent firing buffer in layer II of entorhinal cortex (ECU) with unique temporal context specific representations formed in dentate gyrus (DG), as well as with episodic memory encoded in the CA3 recurrent network. Similarly, spiking representations in DG are associated with episodic memory in CA3. Spike sequences retrieved in CA3 are reversed representations of the original episodic input. As observed by Foster and Wilson, and in more recent electrophysiology, resulting simulated activity in CA1 exhibits some forward and some reversed spiking.

Recent advances in single-neuron biophysics have enhanced our understanding of information processing on the cellular level, but how the detailed properties of individual neurons give rise to large-scale behavior remains unclear. Here, we present a model of the hippocampal network based on observed biophysical properties of hippocampal and entorhinal cortical neurons. We assembled our model to simulate spatial alternation, a task that requires memory of the previous path through the environment for correct selection of the current path to a reward site. The convergence of inputs from entorhinal cortex and hippocampal region CA3 onto CA1 pyramidal cells make them potentially important for integrating information about place and temporal context on the network level. Our model shows how place and temporal context information might be combined in CA1 pyramidal neurons to give rise to splitter cells, which fire selectively based on a combination of place and temporal context. The model leads to a number of experimentally testable predictions that may lead to a better understanding of the biophysical basis of information processing in the hippocampus.

Recent neurophysiological research has focused on the temporal relationships between neuronal firing and plasticity, and has shown the phenomenon of spike-timing-dependent plasticity (STDP). Various models were suggested to implement the STDP-like learning rule in artificial networks based on spiking neuronal representations. Here we present and analyze a simple rule that only depends on the information that is available at the synapse at the time of synaptic modification. This rule is further extended by addition of four different types of gating derived from conventionally used types of gated decay in learning rules for continuous firing rate neural networks. The results show that the advantages of using these gatings are transferred to the new rule without sacrificing its dependency on spike-timing.

Numerous studies have focused on the mechanisms of Hebbian synaptic plasticity within the hippocampal formation and neocortical structures, but the specific role of this plasticity in regulating behavior remains unclean. This paper presents a model demonstrating how Hebbian synaptic modification could mediate the context-dependent retrieval of episodes necessary for memory guided behavior in specific tasks, and could also contribute to prefrontal mechanisms for regulation of goal directed behavior. This modeling effectively links neurophysiological data to the functional role of Hebbian synaptic modification and theta rhythm oscillations. The model simulates data on neuronal spike firing and field potentials in the hippocampus, including: 1) the phenomenon of theta phase precession of place cells: 2) the context-sensitivity of neuronal firing ("splitter cells") during spatial alternation: and 3) the selective changes in theta phase precession across different trials on the same day.

Task specific spiking activity that is selective for specific perceptions and actions is observed in the pre frontal cortex (PFC) of primates and rats during goal-directed behavior. A spiking neuron model of minicolumn circuits in PFC has been shown to successfully replicate the performance and categories of selective neuronal responses recorded in a primate visual discrimination task. The model provides a biological implementation of the action selection process used in reinforcement learning theory. Using this model, we propose a mechanistic explanation based on the reuse of previous encoding in PFC minicolumns for the ability to find short-cuts during the learning of some novel goal-directed tasks, but not others.

The model presented here extends formal analysis (Hasselmo et al., Neural Networks, 15, pp. 689-707, 2002b) and abstract modelling (Gorchetchnikov and Hasselmo, Neurocomputing, 44-46, pp. 423-427, 2002a) of interactions within the hippocampal area (or other cortical areas), which can be flexibly used to navigate toward any arbitrary goal or multiple goals that change on a trial-by-trial basis. The algorithm is a version of a bidirectional breadth-first graph search implemented in simulated neurons using two flows of neural activity. The new model changes the continuous firing rate neuronal representations (Gorchetchnikov and Hasselmo 2002a) to more detailed compartmental versions with realistic parameters, while preserving the qualitative properties analysed previously (Hasselmo et al., 2002b, Gorchetchnikov and Hasselmo 2002a). The case of multiple goals being present in the environment is studied in this paper. The first set of simulations tests the algorithm in the selection of the closest goal. A small difference in distance between the simulated animal and different goals is sufficient for a correct selection. The second set of simulations studies the behaviour of the model when the goals have different saliences. A small salience-based difference between firing rates of the cells providing goal-related input to the model is sufficient for the selection of a more salient goal. This behaviour was tested in three types of environments: a linear track, a T-maze and an open field. Further investigation of quantitative properties of the model should allow it to handle cases when the exact location of the goal is uncertain.

Mechanisms of memory-guided behavior were analyzed in a simulation of the hippocampus and prefrontal cortex which guided movements of a virtual rat in virtual tasks including goal directed movement in the open field, a circular track task, and spatial alternation task. In these tasks, hippocampal circuits mediate retrieval from episodic memory. Possible episodes are retrieved in entorhinal cortex and the interaction with temporal context from dentate gyrus and region CA3 allows selection of specific episodes in CA1. Episodic retrieval allows selection of correct goal-directed movement by prefrontal cortex circuits. Network activity replicates current source density during theta rhythm, as well as theta phase precession and context sensitivity of place cells.

Previous work described a detailed spiking model of interactions between the hippocampus and entorhinal cortex in guiding rat spatial navigation behavior. The timing of spiking activity in the entorhinal cortex is critical for the proper functioning of that model. Here we investigate the influence of several parameters of the model on the spike timing of consecutive traveling pulses within the network of spiking neurons with biologically realistic synaptic parameters. The results suggest that the local properties of the circuit consisting of pyramidal cell and interneuron affect the time interval between consecutive traveling pulses, while the strength of excitatory coupling between these circuits has little influence unless this strength is too small. These results can be generalized to other models of similar architecture that exhibit traveling pulses behavior, and also serve as a basis of further development of full scale realistic implementation of the formal model of spatial navigation.

Memory tasks require processes for encoding associations and retrieving them to guide goal-directed behavior (Hasselmo et al., 2002). Simulations of the interaction of hippocampus and neocortex demonstrate specific dynamical requirements required for encoding episodic memory and learning task rules. 1.) Encoding of new information without interference from previously encoded information requires cyclical transitions between encoding and retrieval states, 2.) Context dependent retrieval of task relevant information from episodic memory requires mechanism for timing and synchronization of inputs from entorhinal cortex and region CA3 to hippocampal region CA1 during retrieval. Theta theory provides a framework for describing how performance is enhanced by specific phase relationships of theta rhythm in different regions of the rat hippocampus and neocortex.

We investigated the importance of hippocampal theta oscillations and the significance of phase differences of theta modulation in the cortical regions that are involved in goal-directed spatial navigation. A model of the rat entorhinal and hippocampal circuitry was created to achieve this. Recurrent fibres between sequentially spiking place cells in ECIII and CA3 established LTP through Hebbian learning that encoded paths through the environment that lead through a goal. Input at arbitrary times during exploration of the environment needed to be repeated as ordered sequences of spikes. A short-term memory (STM) buffer in EC, regulated by theta modulation achieved this and synchronized reactivation with encoding phases in ECIII and CA3. Inhibition at a specific theta phase deactivates the oldest item in the buffer when new input was represented to a full STM buffer. A 180 degree phase difference separated retrieval and encoding in ECIII and CA3, which enabled us to simulate bi-phasic theta phase precession of place cells. Retrieval of known paths was elicited in ECIII by input at the retrieval phase from a PFC storage of goal location. Known locations adjacent to the virtual rat were retrieved in CA3. Together, these activated predictive spiking cells in CA1 for the next desired place on a shortest path to a goal. Consistent with data, place cell activity in CA1 and CA3 showed smaller place fields than in ECIII.

The theta rhythm appears in the rat hippocampal electroencephalogram during exploration and shows phase locking to stimulus acquisition. Lesions that block theta rhythm impair performance in tasks requiring reversal of prior learning, including reversal in a T-maze, where associations between one arm location and food reward need to be extinguished in favor of associations between the opposite arm location and food reward. Here, a hippocampal model shows how theta rhythm could be important for reversal in this task by providing separate functional phases during each 100-300 msec cycle, consistent with physiological data. In the model, effective encoding of new associations occurs in the phase when synaptic input from entorhinal cortex is strong and long-term potentiation (LTP) of excitatory connections arising from hippocampal region CA3 is strong, but synaptic currents arising from region CA3 input are weak (to prevent interference from prior learned associations). Retrieval of old associations occurs in the phase when entorhinal input is weak and synaptic input from region CA3 is strong, but when depotentiation occurs at synapses from CA3 (to allow extinction of prior learned associations that do not match current input). These phasic changes require that LTP at synapses arising from region CA3 should be strongest at the phase when synaptic transmission at these synapses is weakest. Consistent with these requirements, our recent data show that synaptic transmission in stratum radiatum is weakest at the positive peak of local theta, which is when previous data show that induction of LTP is strongest in this layer.

Changes in GABA modulation may underlie experimentally observed changes in the strength of synaptic transmission at different phases of the theta rhythm (Wyble, Linster, & Hasselmo, 1997). Analysis demonstrates that these changes improve sequence disambiguation by a neural network model of CA3. We show that in the framework of Hopfield and Tank (1985), changes in GABA suppression correspond to changes in the effective temperature and the relative energy of data terms and constraints of an analog network. These results suggest that phasic changes in the activity of inhibitory interneurons during a theta cycle may produce dynamics that resemble annealing. These dynamics may underlie a role for the theta cycle in improving sequence retrieval for spatial navigation.

Implementing associative memory function in biologically realistic networks raises difficulties not dealt with in previous associative memory models. In particular, during learning of overlapping input patterns, recall of previously stored patterns can interfere with the learning of new patterns. Most associative memory models avoid this difficulty by ignoring the effect of previously modified connections during learning, thereby clamping activity to the patterns to be learned. Here I propose that the effects of acetylcholine in cortical structures may provide a neuropsychological mechanism for this clamping. Recent brain slice experiments have shown that acetylcholine selectively suppresses excitatory intrinsic fiber synaptic transmission within the olfactory cortex, while leaving excitatory afferent input unaffected. In a computational model of olfactory cortex, this selective suppression, applied during learning, prevents interference from previously stored patterns during the learning of new patterns. Analysis of the model shows that the amount of suppression necessary to prevent interference depends on cortical parameters such as inhibition and the threshold of synaptic modification, as well as input parameters such as the amount of overlap between the patterns being stored.