Neural network architecture for cognitive navigation in dynamic environments

Minimal Neural Network Conditions for Encoding Future Interactions

Space and time are fundamental attributes of the external world. Deciphering the brain mechanisms involved in processing the surrounding environment is one of the main challenges in neuroscience. This is particularly defiant when situations change rapidly over time because of the intertwining of spatial and temporal information. However, understanding the cognitive processes that allow coping with dynamic environments is critical, as the nervous system evolved in them due to the pressure for survival. Recent experiments have revealed a new cognitive mechanism called time compaction. According to it, a dynamic situation is represented internally by a static map of the future interactions between the perceived elements (including the subject itself). The salience of predicted interactions (e.g. collisions) over other spatiotemporal and dynamic attributes during the processing of time-changing situations has been shown in humans, rats, and bats. Motivated by this ubiquity, we study an artificial neural network to explore its minimal conditions necessary to represent a dynamic stimulus through the future interactions present in it. We show that, under general and simple conditions, the neural activity linked to the predicted interactions emerges to encode the perceived dynamic stimulus. Our results show that this encoding improves learning, memorization and decision making when dealing with stimuli with impending interactions compared to no-interaction stimuli. These findings are in agreement with theoretical and experimental results that have supported time compaction as a novel and ubiquitous cognitive process.

Spatial prediction of dynamic interactions in rats

Animals and humans receive the most critical information from parts of the environment that are immediately inaccessible and highly dynamic. The brain must effectively process potential interactions between elements in such an environment to make appropriate decisions in critical situations. We trained male Long-Evans rats to discriminate static and dynamic spatial stimuli and to generalize novel dynamic spatial stimuli displayed on an inaccessible computer screen. We provide behavioral evidence indicating that rats encode dynamic visuospatial situations by constructing internal static representations that capture meaningful future interactions between objects. These observations support previous findings in humans that such internal static representations can encapsulate relevant spatiotemporal information of dynamic environments. This mechanism would allow animals and humans to process complex time-changing situations neatly.

Enhancing Underwater SLAM Navigation and Perception: A Comprehensive Review of Deep Learning Integration

Underwater simultaneous localization and mapping (SLAM) is essential for effectively navigating and mapping underwater environments; however, traditional SLAM systems have limitations due to restricted vision and the constantly changing conditions of the underwater environment. This study thoroughly examined the underwater SLAM technology, particularly emphasizing the incorporation of deep learning methods to improve performance. We analyzed the advancements made in underwater SLAM algorithms. We explored the principles behind SLAM and deep learning techniques, examining how these methods tackle the specific difficulties encountered in underwater environments. The main contributions of this work are a thorough assessment of the research into the use of deep learning in underwater image processing and perception and a comparison study of standard and deep learning-based SLAM systems. This paper emphasizes specific deep learning techniques, including generative adversarial networks (GANs), convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and other advanced methods to enhance feature extraction, data fusion, scene understanding, etc. This study highlights the potential of deep learning in overcoming the constraints of traditional underwater SLAM methods, providing fresh opportunities for exploration and industrial use.

Detecting and Tracking of Multiple Mice Using Part Proposal Networks

The study of mouse social behaviors has been increasingly undertaken in neuroscience research. However, automated quantification of mouse behaviors from the videos of interacting mice is still a challenging problem, where object tracking plays a key role in locating mice in their living spaces. Artificial markers are often applied for multiple mice tracking, which are intrusive and consequently interfere with the movements of mice in a dynamic environment. In this article, we propose a novel method to continuously track several mice and individual parts without requiring any specific tagging. First, we propose an efficient and robust deep-learning-based mouse part detection scheme to generate part candidates. Subsequently, we propose a novel Bayesian-inference integer linear programming (BILP) model that jointly assigns the part candidates to individual targets with necessary geometric constraints while establishing pair-wise association between the detected parts. There is no publicly available dataset in the research community that provides a quantitative test bed for part detection and tracking of multiple mice, and we here introduce a new challenging Multi-Mice PartsTrack dataset that is made of complex behaviors. Finally, we evaluate our proposed approach against several baselines on our new datasets, where the results show that our method outperforms the other state-of-the-art approaches in terms of accuracy. We also demonstrate the generalization ability of the proposed approach on tracking zebra and locust.

Unsupervised Estimation of Monocular Depth and VO in Dynamic Environments via Hybrid Masks

Deep learning-based methods mymargin have achieved remarkable performance in 3-D sensing since they perceive environments in a biologically inspired manner. Nevertheless, the existing approaches trained by monocular sequences are still prone to fail in dynamic environments. In this work, we mitigate the negative influence of dynamic environments on the joint estimation of depth and visual odometry (VO) through hybrid masks. Since both the VO estimation and view reconstruction process in the joint estimation framework is vulnerable to dynamic environments, we propose the cover mask and the filter mask to alleviate the adverse effects, respectively. As the depth and VO estimation are tightly coupled during training, the improved VO estimation promotes depth estimation as well. Besides, a depth-pose consistency loss is proposed to overcome the scale inconsistency between different training samples of monocular sequences. Experimental results show that both our depth prediction and globally consistent VO estimation are state of the art when evaluated on the KITTI benchmark. We evaluate our depth prediction model on the Make3D dataset to prove the transferability of our method as well.

Benchmarking of tools for axon length measurement in individually-labeled projection neurons

Projection neurons are the commonest neuronal type in the mammalian forebrain and their individual characterization is a crucial step to understand how neural circuitry operates. These cells have an axon whose arborizations extend over long distances, branching in complex patterns and/or in multiple brain regions. Axon length is a principal estimate of the functional impact of the neuron, as it directly correlates with the number of synapses formed by the axon in its target regions; however, its measurement by direct 3D axonal tracing is a slow and labor-intensive method. On the contrary, axon length estimations have been recently proposed as an effective and accessible alternative, allowing a fast approach to the functional significance of the single neuron. Here, we analyze the accuracy and efficiency of the most used length estimation tools—design-based stereology by virtual planes or spheres, and mathematical correction of the 2D projected-axon length—in contrast with direct measurement, to quantify individual axon length. To this end, we computationally simulated each tool, applied them over a dataset of 951 3D-reconstructed axons (from NeuroMorpho.org), and compared the generated length values with their 3D reconstruction counterparts. The evaluated reliability of each axon length estimation method was then balanced with the required human effort, experience and know-how, and economic affordability. Subsequently, computational results were contrasted with measurements performed on actual brain tissue sections. We show that the plane-based stereological method balances acceptable errors (~5%) with robustness to biases, whereas the projection-based method, despite its accuracy, is prone to inherent biases when implemented in the laboratory. This work, therefore, aims to provide a constructive benchmark to help guide the selection of the most efficient method for measuring specific axonal morphologies according to the particular circumstances of the conducted research.

Characterization of single neurons is a crucial step to understand how neural circuitry operates. Visualization of individual neurons is feasible thanks to labelling techniques that allow precise measurements at cellular resolution. This milestone gave access to powerful estimators of the functional impact of a neuron, such as axon length. Although techniques relying on direct 3D reconstruction of individual axons are the gold standard, handiness and accessibility are still an issue. Indirect estimations of axon length have been proposed as agile and effective alternatives, each offering different solutions to the accuracy-cost tradeoff. In this work we report a computational benchmarking between three experimental tools used for axon length estimation on brain tissue sections. Performance of each tool was simulated and tested for 951 3D-reconstructed axons, by comparing estimated axon lengths against direct measurements. Assessment of suitability to different research and funding circumstances is also provided, taking into consideration factors such as training expertise, economic cost and required equipment, alongside methodological results. These findings could be an important reference for research on neuronal wiring, as well as for broader studies involving neuroanatomical and neural circuit modelling.

Considering Neural Connectivity in Point Process Decoder for Brain-Machine Interface. ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY. IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY. ANNUAL INTERNATIONAL CONFERENCE 2021;2021:6341-6344. [PMID: 34892563 DOI: 10.1109/embc46164.2021.9630383] [Citation(s) in RCA: 0] [Impact Index Per Article: 0]

Brain machine interface (BMI) can translate neural activity into digital commands to control prostheses. The decoder in BMI models the mechanism relating to neural activity and intents in brain. In our brain, single neuronal tuning property and neural connectivity contribute to encoding the intents together. These properties may change, a phenomenon which is named neural adaptation during using BMIs. Neural adaptation requires the decoder to consider the two factors at the same time and has the potential to follow their changes. However, in the previous work, the class of neural network and clustering decoder can consider the neural connectivity regardless of the single neuronal tuning property. On the other hand, point process methods can model the single neuronal tuning property but fail to address the neural connectivity. In this paper, we propose a new point process decoder with the information of neural connectivity named NCPP. We derive the neural connectivity component from the point process method by Bayes' rule and use a clustering decoder to represent the neural connectivity. This method can consider the neural connectivity and the single neuronal tuning property at the same time. We validate the method on simulation data where the point process method cannot achieve a good decoding performance and compare it with sequential Monte Carlo point process method (SMCPP). The results show our method outperforms the pure point process method which indicates our method can model the neural connectivity and single neuronal tuning property at the same time.Clinical Relevance-This paper proposes a decoder that can model the neural connectivity and the single neuronal tuning property at the same time, which is potential to explain the neural adaptation computationally.

Spatial Memory in a Spiking Neural Network with Robot Embodiment

Cognitive maps and spatial memory are fundamental paradigms of brain functioning. Here, we present a spiking neural network (SNN) capable of generating an internal representation of the external environment and implementing spatial memory. The SNN initially has a non-specific architecture, which is then shaped by Hebbian-type synaptic plasticity. The network receives stimuli at specific loci, while the memory retrieval operates as a functional SNN response in the form of population bursts. The SNN function is explored through its embodiment in a robot moving in an arena with safe and dangerous zones. We propose a measure of the global network memory using the synaptic vector field approach to validate results and calculate information characteristics, including learning curves. We show that after training, the SNN can effectively control the robot’s cognitive behavior, allowing it to avoid dangerous regions in the arena. However, the learning is not perfect. The robot eventually visits dangerous areas. Such behavior, also observed in animals, enables relearning in time-evolving environments. If a dangerous zone moves into another place, the SNN remaps positive and negative areas, allowing escaping the catastrophic interference phenomenon known for some AI architectures. Thus, the robot adapts to changing world.

Static internal representation of dynamic situations reveals time compaction in human cognition

Introduction

The human brain has evolved under the constraint of survival in complex dynamic situations. It makes fast and reliable decisions based on internal representations of the environment. Whereas neural mechanisms involved in the internal representation of space are becoming known, entire spatiotemporal cognition remains a challenge. Growing experimental evidence suggests that brain mechanisms devoted to spatial cognition may also participate in spatiotemporal information processing.

Objectives

The time compaction hypothesis postulates that the brain represents both static and dynamic situations as purely static maps. Such an internal reduction of the external complexity allows humans to process time-changing situations in real-time efficiently. According to time compaction, there may be a deep inner similarity between the representation of conventional static and dynamic visual stimuli. Here, we test the hypothesis and report the first experimental evidence of time compaction in humans.

Methods

We engaged human subjects in a discrimination-learning task consisting in the classification of static and dynamic visual stimuli. When there was a hidden correspondence between static and dynamic stimuli due to time compaction, the learning performance was expected to be modulated. We studied such a modulation experimentally and by a computational model.

Results

The collected data validated the predicted learning modulation and confirmed that time compaction is a salient cognitive strategy adopted by the human brain to process time-changing situations. Mathematical modelling supported the finding. We also revealed that men are more prone to exploit time compaction in accordance with the context of the hypothesis as a cognitive basis for survival.

Conclusions

The static internal representation of dynamic situations is a human cognitive mechanism involved in decision-making and strategy planning to cope with time-changing environments. The finding opens a new venue to understand how humans efficiently interact with our dynamic world and thrive in nature.

Semantic Knowledge Representation for Strategic Interactions in Dynamic Situations

Evolved living beings can anticipate the consequences of their actions in complex multilevel dynamic situations. This ability relies on abstracting the meaning of an action. The underlying brain mechanisms of such semantic processing of information are poorly understood. Here we show how our novel concept, known as time compaction, provides a natural way of representing semantic knowledge of actions in time-changing situations. As a testbed, we model a fencing scenario with a subject deciding between attack and defense strategies. The semantic content of each action in terms of lethality, versatility, and imminence is then structured as a spatial (static) map representing a particular fencing (dynamic) situation. The model allows deploying a variety of cognitive strategies in a fast and reliable way. We validate the approach in virtual reality and by using a real humanoid robot.

Adaptive-Cognitive Kalman Filter and Neural Network for an Upgraded Nondispersive Thermopile Device to Detect and Analyze Fusarium Spores

Noises such as thermal noise, background noise or burst noise can reduce the reliability and confidence of measurement devices. In this work, a recursive and adaptive Kalman filter is proposed to detect and process burst noise or outliers and thermal noise, which are popular in electrical and electronic devices. The Kalman filter and neural network are used to preprocess data of three detectors of a nondispersive thermopile device, which is used to detect and quantify Fusarium spores. The detectors are broadband (1 µm to 20 µm), λ 1 (6.09 ± 0.06 µm) and λ 2 (9.49 ± 0.44 µm) thermopiles. Additionally, an artificial neural network (NN) is applied to process background noise effects. The adaptive and cognitive Kalman Filter helps to improve the training time of the neural network and the absolute error of the thermopile data. Without applying the Kalman filter for λ 1 thermopile, it took 12 min 09 s to train the NN and reach the absolute error of 2.7453 × 104 (n. u.). With the Kalman filter, it took 46 s to train the NN to reach the absolute error of 1.4374 × 104 (n. u.) for λ 1 thermopile. Similarly, to the λ 2 (9.49 ± 0.44 µm) thermopile, the training improved from 9 min 13 s to 1 min and the absolute error of 2.3999 × 105 (n. u.) to the absolute error of 1.76485 × 105 (n. u.) respectively. The three-thermopile system has proven that it can improve the reliability in detection of Fusarium spores by adding the broadband thermopile. The method developed in this work can be employed for devices that encounter similar noise problems.

High-Dimensional Brain: A Tool for Encoding and Rapid Learning of Memories by Single Neurons

Codifying memories is one of the fundamental problems of modern Neuroscience. The functional mechanisms behind this phenomenon remain largely unknown. Experimental evidence suggests that some of the memory functions are performed by stratified brain structures such as the hippocampus. In this particular case, single neurons in the CA1 region receive a highly multidimensional input from the CA3 area, which is a hub for information processing. We thus assess the implication of the abundance of neuronal signalling routes converging onto single cells on the information processing. We show that single neurons can selectively detect and learn arbitrary information items, given that they operate in high dimensions. The argument is based on stochastic separation theorems and the concentration of measure phenomena. We demonstrate that a simple enough functional neuronal model is capable of explaining: (i) the extreme selectivity of single neurons to the information content, (ii) simultaneous separation of several uncorrelated stimuli or informational items from a large set, and (iii) dynamic learning of new items by associating them with already "known" ones. These results constitute a basis for organization of complex memories in ensembles of single neurons. Moreover, they show that no a priori assumptions on the structural organization of neuronal ensembles are necessary for explaining basic concepts of static and dynamic memories.

Fast social-like learning of complex behaviors based on motor motifs

Social learning is widely observed in many species. Less experienced agents copy successful behaviors exhibited by more experienced individuals. Nevertheless, the dynamical mechanisms behind this process remain largely unknown. Here we assume that a complex behavior can be decomposed into a sequence of n motor motifs. Then a neural network capable of activating motor motifs in a given sequence can drive an agent. To account for (n-1)! possible sequences of motifs in a neural network, we employ the winnerless competition approach. We then consider a teacher-learner situation: one agent exhibits a complex movement, while another one aims at mimicking the teacher's behavior. Despite the huge variety of possible motif sequences we show that the learner, equipped with the provided learning model, can rewire "on the fly" its synaptic couplings in no more than (n-1) learning cycles and converge exponentially to the durations of the teacher's motifs. We validate the learning model on mobile robots. Experimental results show that the learner is indeed capable of copying the teacher's behavior composed of six motor motifs in a few learning cycles. The reported mechanism of learning is general and can be used for replicating different functions, including, for example, sound patterns or speech.

Quantized Attention-Gated Kernel Reinforcement Learning for Brain-Machine Interface Decoding

Reinforcement learning (RL)-based decoders in brain-machine interfaces (BMIs) interpret dynamic neural activity without patients' real limb movements. In conventional RL, the goal state is selected by the user or defined by the physics of the problem, and the decoder finds an optimal policy essentially by assigning credit over time, which is normally very time-consuming. However, BMI tasks require finding a good policy in very few trials, which impose a limit on the complexity of the tasks that can be learned before the animal quits. Therefore, this paper explores the possibility of letting the agent infer potential goals through actions over space with multiple objects, using the instantaneous reward to assign credit spatially. A previous method, attention-gated RL employs a multilayer perceptron trained with backpropagation, but it is prone to local minima entrapment. We propose a quantized attention-gated kernel RL (QAGKRL) to avoid the local minima adaptation in spatial credit assignment and sparsify the network topology. The experimental results show that the QAGKRL achieves higher successful rates and more stable performance, indicating its powerful decoding ability for more sophisticated BMI tasks as required in clinical applications.

An Efficient Fine-to-Coarse Wayfinding Strategy for Robot Navigation in Regionalized Environments

This paper proposes an efficient wayfinding strategy for robot navigation in regionalized environments by designing a regionalized spatial knowledge model (RSK model) and a region-based wayfinding algorithm, i.e., a fine-to-coarse A* (FTC-A*) search algorithm. First, the RSK model, which imitates the representation of environments in the human brain, is presented to describe the search environments. The environments that are divided into regions are represented by a hierarchical nested structure where small regions are grouped together to form superordinate regions. Second, on the basis of the RSK model, an FTC-A* search algorithm is developed to plan the fine-to-coarse route. By making a fine planning to robot surroundings in vicinity, but a coarse planning to that at the distance, the FTC-A* algorithm can effectively reduce computational complexity, so as to enhance the efficiency of route search, and meanwhile makes robots to react quickly to user's commands, especially in large-scale environments. Finally, four exhaustive simulations and a physical experiment have been carried out to illustrate the feasibility and effectiveness of the proposed wayfinding strategy.

Prediction-for-CompAction: navigation in social environments using generalized cognitive maps

The ultimate navigation efficiency of mobile robots in human environments will depend on how we will appraise them: merely as impersonal machines or as human-like agents. In the latter case, an agent may take advantage of the cooperative collision avoidance, given that it possesses recursive cognition, i.e., the agent's decisions depend on the decisions made by humans that in turn depend on the agent's decisions. To deal with this high-level cognitive skill, we propose a neural network architecture implementing Prediction-for-CompAction paradigm. The network predicts possible human-agent collisions and compacts the time dimension by projecting a given dynamic situation into a static map. Thereby emerging compact cognitive map can be readily used as a "dynamic GPS" for planning actions or mental evaluation of the convenience of cooperation in a given context. We provide numerical evidence that cooperation yields additional room for more efficient navigation in cluttered pedestrian flows, and the agent can choose path to the target significantly shorter than a robot treated by humans as a functional machine. Moreover, the navigation safety, i.e., the chances to avoid accidental collisions, increases under cooperation. Remarkably, these benefits yield no additional load to the mean society effort. Thus, the proposed strategy is socially compliant, and the humanoid agent can behave as "one of us."