Neural Networks

Wave recurrent neural networks (wRNNs) are biologically inspired recurrent architectures that use traveling-wave dynamics to support sequence learning and memory. However, their input-to-hidden pathway remains relatively simple compared with biological neurons, where dendrites perform nonlinear input integration. In this study, we introduce the Dendritic Wave Recurrent Neural Network (DWRNN), which augments the input pathway of the wRNN with nonlinear basal dendritic branches while preserving the original recurrent wave dynamics. We evaluate DW-RNN on a simple copy task, sequential MNIST (sMNIST), permuted sequential MNIST (psMNIST), and noisy sequential CIFAR-10 (nsCIFAR-10). On the copy task, DW-RNN shows learning behavior comparable to the standard wRNN, suggesting that dendritic input integration does not disrupt the recurrent wave-based memory mechanism. On the three sequential image-classification benchmarks, DW-RNN outperforms the standard wRNN, improving accuracy from 97.27 {+/-} 0.15% to 97.82 {+/-} 0.12% on sMNIST, from 96.74 {+/-} 0.17% to 96.92 {+/-} 0.10% on psMNIST, and from 54.30 {+/-} 0.79% to 55.65 {+/-} 0.55% on nsCIFAR-10. In addition to improving mean accuracy, DW-RNN exhibits lower across-seed variability on all three classification benchmarks, suggesting that dendritic input integration may improve the stability of wRNN training. Hidden-activity visualizations further show that DW-RNN preserves the characteristic traveling-wave patterns of the original wRNN. These results suggest that dendritic computation and traveling-wave recurrent dynamics provide complementary mechanisms for biologically inspired sequence learning.

Equilibrium propagation (EP) is a biologically plausible alternative to backpropagation for training neural networks. EP typically relies on free and nudged dynamical phases: during the free phase, the network relaxes toward an equilibrium state, whereas during nudging, the output state is perturbed toward a target using a teaching signal. However, it remains unclear whether the brain has access to such explicit target signals. Inspired by Attention-Gated Brain Propagation (BrainProp), a reward-based learning framework proposed by Pozzi et al. (2020), we introduce a reward-based variant of EP that replaces full target-based nudging with a selected-output binary reward signal. The proposed method updates the network using only the chosen class and whether that choice is correct, without directly revealing the full target vector. We evaluate the method on MNIST, Fashion-MNIST, and CIFAR-10 using both multilayer perceptrons and convolutional neural networks. The proposed reward-based EP achieves performance close to that of conventional EP across all three datasets, although it generally converges more slowly during the early stages of training. Generalization-gap analyses show similar behavior for the two methods on MNIST and Fashion-MNIST, while reward-based EP exhibits a smaller training-test accuracy gap during later training on CIFAR-10. We further investigate the effect of the exploration probability used during stochastic class selection and find that moderate exploration can provide small performance improvements, although its effect is dataset-dependent. These results demonstrate that EP can learn effectively from sparse, action-specific reward feedback rather than a complete supervised target.

Cerebral cortical networks in the mammalian brain exhibit a non-random organization in which reciprocal projections, although widespread, are systematically asymmetric in strength: feedforward connections are consistently stronger than their feedback counterparts, particularly in sensory cortices. This "no-strong-loops" principle is thought to prevent runaway excitation and maintain stability, yet its actual computational impact remains unclear. Here, we use computational analysis and modeling to show that connectivity asymmetry supports high working-memory capacity, whereas increasing reciprocity reduces memory capacity and representational diversity in reservoir-computing models of recurrent neural networks. We systematically examine synthetic architectures inspired by mammalian cortical connectivity and find that sparse, modular, and hierarchical networks achieve superior performance, relative to random, small-world, or core-periphery graphs, but only when reciprocity is constrained. Validated on directed mammalian (macaque, marmoset, rat, and mouse) connectomes, these results indicate that restricting reciprocal motifs yields functional benefits in sparse networks, consistent with an evolutionary strategy for stable, efficient information processing in the brain. These findings suggest a biologically-inspired design principle for artificial neural systems.

The genomic mechanisms that efficiently encode the initial architecture and synaptic connectivity of neural circuits remain poorly understood. We hypothesise that two primary mechanisms -- spatial encoding and factorisation -- enable a limited genome to initialise networks of billions of neurons. Spatial encoding, a form of indirect representation, compresses neural network parameters while enforcing structural continuity. Complementarily, we introduce a factorisation mechanism inspired by reaction-diffusion models, comprising a spatially invariant reaction rule and spatially variant diffusion dynamics. Based on this insight, we can efficiently abstract a neural network layer as a spatially invariant weight kernel (or filter) and its spatially variant transformations along the spatial dimensions. Thus, coupling this variant-invariant decomposition with spatial encodings can substantially reduce the size of the solution space explored by the genome. In addition, we show that this coupling leads to efficient initialisation of cortical maps, such as V1 orientation maps, and neural networks with the ability to generalise. In summary, coupling factorisation and spatial encodings can offer functional advantages to the evolving genome.

Animals must often navigate environments where feedback about progress toward a goal is sparse or delayed, requiring internal representations of space and memory of prior experience. The hippocampal-entorhinal system is believed to support this capability through distributed spatial representations that guide goal-directed behavior. However, many computational models of these circuits focus primarily on reproducing neural dynamics rather than demonstrating how such representations support learning on navigation tasks. We present a biologically inspired spiking neuronal network (SNN) model that combines grid-cell-derived spatial representations, {Delta}Q-modulated Hebbian plasticity, and context-dependent modulation to support navigation under sparse reward conditions. Grid Cell populations generate distributed spatial codes that are transformed by an Association Cell population into more spatially selective internal representations. Learning is driven by changes in Q-values ({Delta}Q) computed from a goal-conditioned Q-table, allowing local synaptic plasticity to incorporate information about long-term navigation outcomes. For environments containing multiple navigation objectives, a Context Cell population provides task-dependent modulation that enables a shared network architecture to support distinct navigation policies. Across two complementary maze environments, the model demonstrates three core capabilities: generation of distinct spatial representations, learning of efficient navigation policies under sparse and delayed reward, and support for multiple navigation objectives within a shared environment. The results further show that contextual modulation introduces subtle task-dependent variations into a largely shared population representation, allowing identical spatial locations to support different navigation behaviors. These findings demonstrate that biologically inspired spatial representations, value-guided plasticity, and contextual modulation can jointly support flexible navigation in spiking neuronal networks, providing a bridge between mechanistic neural circuit models and functional reinforcement learning.

Recurrent neural networks (RNNs) trained with backpropagation through time (BPTT) use gradients and global error signals to solve tasks, while evolutionary algorithms (EAs) offer an alternative solution through gradient-free optimization. Both classes of methods can solve the same tasks by modifying network weights, but it remains unclear how the choice of training paradigm biases the final connectivity structure and resulting dynamics. When drawing conclusions from task-trained RNNs, especially as proxies for neurobiological computation, it is important to consider whether the resulting network structure is due to the training method itself. Here, we compare four training paradigms--BPTT, evolution strategies (ES), genetic algorithms (GA), and GA combined with Ojas Hebbian plasticity rule (GA+Oja) -- across RNNs of 32, 64, and 128 neurons. These training paradigms were applied to two contrasting tasks that are exemplary of tasks which animals perform: a discrete working memory task and a continuous sensorimotor integration task. For both tasks, we analyzed how each training algorithm modified the network through four measurements: weight change allocations across input (Win), recurrent (Wrec), and output (Wout) layers, the effective rank of the recurrent weight matrix, the dimensionality of hidden-state dynamics, and task accuracy. BPTT achieved near-perfect accuracy across all task conditions, progressively allocated more weight changes to Wout as the working memory tasks difficulty increased, and confined hidden-state activity to a lower-dimensional subspace than any evolutionary method. Evolutionary methods maintained higher recurrent effective rank, higher activity dimensionality, and no comparable difficulty-dependent reallocation toward the readout, all while maintaining comparable task accuracy as BPTT. These findings show how gradient-based and gradient-free algorithms discover distinct structural and dynamical solutions to the same computational problems, especially in tasks involving working memory, with important implications for analyzing task-trained RNNs as models of biological neural computation. Biological neural circuits, which are shaped by evolution and local plasticity rather than gradient descent, may operate in higher-dimensional regimes than gradient-trained RNN models predict.

Recurrence is thought to enhance the robustness of biological vision, but how it achieves this feat is largely unknown. Perceptual robustness can be implemented through either lateral connections supporting local integration within a processing stage or feedback connections drawing on broader context from higher stages, and through either a discriminative objective optimising task-relevant classification or a generative objective learning to reconstruct the causes of visual input. But do these different types of recurrence engage distinct computational strategies? As this question is difficult to test in vivo, we endowed convolutional neural networks with varying recurrent architectures and training objectives, and evaluated the consequences for internal representations and behaviour across noise levels. Two distinct computational strategies emerged. Generative feedback followed a reductionist strategy, with representations becoming lower-dimensional through denoising, achieving robustness at moderate noise levels without noise training. Both discriminative lateral and feedback recurrence followed an expansionist strategy, increasing dimensionality to sharpen discriminability without denoising, but requiring noise training to achieve robustness. These dissociable signatures reflect fundamentally different computational mechanisms of robust vision and provide testable predictions for which form of recurrence the brain employs.

Robotic motor control built on liquid neural networks and related continuous-time models, such as LTC and CfC, is typically trained offline via backpropagation through time and lacks an explicit mechanism for recalibrating online as plant dynamics change. We ask whether a frozen CfC core, whose liquid state spans a fixed continuous-time basis, can support cerebellar-style online adaptation by adapting only its linear readout with a climbing-fiber-like error signal. In a planar two-link reaching simulation with a velocity-dependent curl force field, we adapt the readout online with a feedback-error-learning (FEL) signal under a least-mean-squares (LMS) rule, leaving the core untouched. The frozen-core readout-only controller re-straightens curl-perturbed reaches and, upon field removal, produces a mirror-image after-effect, a behavioral signature consistent with internal-model learning, which a feedback-only controller does not produce. The result generalizes from a dense CfC to a sparse Neural-Circuit-Policy (NCP) wiring when the recurrent state, rather than the projected motor output, is used as the readout basis; it is robust to force-field strength and direction; and a recursive-least-squares variant adapts faster but de-adapts slowly because its covariance collapses, a rigidity that a covariance-reset safe-forgetting rule removes. Within the explored two-link planar simulation range, we did not find a readout-only failure case that required adapting the frozen core in the tested conditions. In this simulation study, adapting only the readout therefore provides a biologically inspired, low-cost online error-adaptation layer for offline-trained continuous-time controllers.

Abnormal states such as erroneous behaviors are generally difficult to represent from neural data. However, such states are also known to have specific spatiotemporal features, indicating a feasibility of developing a method to focus on them. If a method can highlight these spatiotemporal features, it may effectively represent such abnormal states, helping evaluate abnormal brain functions. In the present study, we proposed the hierarchy of supported modules (HSM) to highlight spatiotemporal features that can represent abnormal states. HSM spatiotemporally transforms multidimensional neural time-series based on their spatiotemporal context. We evaluated HSM through decoding and similarity analyses using multiple publicly available datasets. In the HSM results, decoding accuracies were higher for erroneous behaviors than for normal behaviors, and similarities were lower between erroneous behaviors and normal behaviors than between normal behaviors, demonstrating the ability of HSM to capture the spatiotemporal features of erroneous behaviors. Surprisingly, many parts of these results were also present even before HSM learning, showing the virtue of HSM as a simple-to-use method. The proposed HSM method may help elucidate the mechanisms underlying erroneous behaviors.

Orientation selectivity in the visual cortex has been extensively studied since it was disovered in 1959. In this article we demonstrate that current mechanistic explanations of orientation selectivity in the visual cortex, which appear to work in simplified models, do not suffice when embedded in a more realistic network. Up to now, orientation selectivity has only been reproduced in models with distance-dependent connectivity and limited cellular diversity. We therefore constructed a mouse primary visual cortex model with, among other refinements, realistic neuronal diversity and morphologically constrained connectivity. We implemented proposed mechanisms of orientation selectivity - specifically the organization of thalamocortical input into elliptical subfields and peferential connectivity between neurons with similar receptive fields - within this model. In the process we explored limitations of pure anatomy in predicting the connectivity and function of the cortex. Simulations of our model did not reproduce realistic levels of orientation selectivy. This indicates that these mechanisms, despite working well in simpler models, are not sufficient to explain orientation selectivity in real brain networks. On the basis of our results, we suggest the hypothesis that homeostatic plasticity is necessary to reconcile orientation selectivity with realistic connectivity.

Medical imaging pipelines routinely copy single-channel grayscale data into three identical RGB channels before classification, usually without justification. This study tests whether that step affects model predictions. Four coordinated experiments on bit-identical RGB inputs sorted eleven classical machine learning models into three groups: five that were invariant to the copy, two that were nearly invariant, and four whose predictions changed. On the Kaggle Alzheimer MRI Dataset (6,400 images, four classes, five seeds), five models (AdaBoost, HistGradientBoosting, KNN, SVM_Polynomial, and SVM_RBF) produced identical predictions in both conditions for every seed, where KNN is k-nearest neighbors and SVM a support vector machine, with polynomial and radial basis function (RBF) kernels. Two models (GaussianNB and SVM_Linear) differed by at most one of 1,280 samples, a dataset-dependent gap rather than exact invariance. The remaining four (DecisionTree, ExtraTrees, RandomForest, and LogisticRegression) differed substantively. A regularization sweep on Logistic Regression traced its gap to a single cause. As L2 regularization weakened, the color-minus-grayscale macro F1 gap shrank steadily, from +12.07 percentage points at C=0.001 to near zero at C=100 (paired Wilcoxon p=0.0020 under strong regularization), showing the effect scales with feature count rather than image content. A replication on the OASIS dataset, matched in size and class balance, reproduced every grouping, and the Logistic Regression gap reappeared in the same direction at smaller magnitude (+5.30 points macro F1). Two deep networks, ResNet18 and DenseNet121, gave identical predictions across all twenty paired conditions. Channel triplication left most models unchanged while multiplying classical training time 2.3 to 4.0 times without benefit.

Spike-timing-dependent plasticity (STDP) and dopamine (DA) are fundamental to reward-based learning and memory formation. A widely used DA-modulated STDP model explains how neural networks associate stimuli with delayed dopaminergic rewards through an eligibility trace. However, we show that this model supports learning even at unrealistically high DA concentrations because DA simply scales the magnitude of STDP without changing its temporal profile. In contrast, experiments demonstrate that DA nonlinearly reshapes the STDP window, converting long-term depression (LTD) into long-term potentiation (LTP) at high DA levels. We therefore propose a DA-modulated STDP rule in which increasing DA progressively biases plasticity toward potentiation while receptor saturation limits further DA effects beyond a critical concentration. Simulations of recurrent networks of Izhikevich neurons show that the proposed rule supports robust conditioning only within a biologically realistic DA range (0.04-0.70 {micro}M). Successful learning produces a hybrid network architecture consisting of a strong feedforward backbone embedded within recurrent circuitry and generates enhanced burst responses selectively to reward-associated stimuli. At the upper limit of the biologically plausible DA range, the network passes through a narrow bistable regime, converging to one of two distinct stable configurations. At higher DA concentrations, conditioning fails altogether. These results provide a biologically grounded model of DA-dependent plasticity and offer new insight into how abnormal dopamine signaling can impair learning in neurological disorders.

Background / Introduction: Biologically inspired spiking neural networks can model adaptive behavior, but learning multiple goals is difficult because synaptic updates for different targets can interfere. We tested whether multi-timescale plasticity and context-specific credit assignment could improve continual multi-goal learning in a spiking navigation system inspired by entorhinal-hippocampal circuitry. Methods: We developed a closed-loop spiking model containing grid-like, place-like, target-related, association, and motor-output populations. An agent navigated in a two-dimensional environment with randomized starting locations and learned through reward-modulated spike-timing dependent plasticity (STDP/RL) and a novel evidence-gated plasticity (EGP) framework. EGP accumulates candidate synaptic modifications, evaluates them using reward evidence, and consolidates only changes that improve performance. A target-context variant maintained separate proposal stores and reward evaluation for each target. Results: STDP/RL learned and retained a single-target navigation policy, but multi-target training produced substantial interference, including attraction to incorrect targets after learning. Across 10 connectivity seeds, target-context EGP achieved higher late-stage reward than global EGP, improved weakest-target performance, and increased the fraction of targets achieving positive reward. In a longer continual-learning simulation, reward increased for all targets, TEST-phase performance increasingly exceeded TRAIN-phase performance, and proposal magnitudes grew over learning. Dwell-time confusion analyses showed that target-context EGP reduced wrong-target attraction and improved target selectivity relative to multi-target STDP/RL. Conclusions: These results demonstrate that spiking navigation circuits can learn goal-directed behavior using local plasticity, but robust multi-goal learning benefits from context-specific evidence-based consolidation. Target-context EGP provides a biologically motivated mechanism for reducing interference during continual reinforcement learning in spiking neural networks.

AO_SCPLOWBSTRACTC_SCPLOWHigh-dimensional neural activity often resides in a low-dimensional subspace, referred to as neural manifolds. Grid cells in the medial entorhinal cortex provide a periodic spatial code that is organized near a toroidal manifold, independent of the spatial environment. Due to the periodic nature of this code, it is unclear how the brain utilizes the toroidal manifold to understand its state in a spatial environment. We introduce a novel framework that decodes spatial information from grid cell activity using topology. Our approach uses topological data analysis to extract toroidal coordinates from grid cell population activity and employs path-lifting to reconstruct trajectories in physical space. The reconstructed paths differ from the original by an affine transformation. We validated the method on both continuous attractor network simulations and experimental recordings of grid cells, demonstrating that local trajectories can be reliably reconstructed from a single grid cell module without external position information or training data. These results suggest that co-modular grid cells contain sufficient information for path integration and suggest a potential computational mechanism for spatial navigation.

Flexibly recombining computational modules is essential for biological and artificial neural networks to rapidly adapt to changing environments. This requires modules to be shared across tasks rather than rigidly segregated, yet what determines this organization remains unknown. Previous work suggests that weight initialization shapes whether networks learn task-specific or generic representations, but it is unclear whether this extends to recurrent networks and, more importantly, to network connectivity. Here, we systematically vary the initial weight variance of recurrent neural networks and study them using a framework that allows us to identify the functionally relevant connectivity subspaces for each computational module. We find that networks with low initial weight variance converge to solutions in which different subtasks rely on largely overlapping weight subspaces, whereas high-variance networks implement subtasks in higher-dimensional, more segregated weight subspaces. Our results also provide mechanistic insights with implications for interpreting biological neural circuits and for designing efficient recurrent architectures.

The ability to extract and exploit temporal structure across diverse tasks is central to human cognition. Neuroscientists have typically relied on recurrent neural networks (RNNs) trained with backpropagation through time (BPTT) when modelling neural and behavioural processes such as decision-making and motor control. However, this algorithm has limited biological plausibility, hence the computational principles underlying efficient learning of temporal dependencies remain unresolved. Here, we investigate temporal predictive coding (tPC), a recently proposed framework that extends predictive coding to the temporal domain while preserving local Hebbian update rules. We analyse and extend tPC to establish its relationship with several influential computational models of learning in RNNs, including BPTT, reservoir computing, and eligibility propagation (e-prop). We first demonstrate a functional equivalence between tPC and tBPTT1, a variant of BPTT in which gradients are propagated only one time step into the past. We then show that tPC can leverage reservoir dynamics to encode short-range temporal context, and simultaneously sculpt neural trajectories in state space to support downstream readout. We further demonstrate that hierarchical recurrent dynamics can facilitate learning of more complex temporal dependencies, while additionally conferring robustness to strong distractors. Finally, we show that tPC networks can be augmented with biologically inspired eligibility traces to solve temporally extended context-dependent tasks. Together, these results reveal that relatively simple recurrent networks governed by local plasticity can support temporal learning in more complex settings than previously appreciated.

Stabilizing synaptic plasticity together with the neurons activity has remained a central challenge in theoretical neuroscience since the introduction of Hebbian learning principles. Classical Hebbian learning rules typically lead to unbounded synaptic growth, motivating the development of stabilization mechanisms such as normalization methods, BCM-type learning, synaptic scaling and others. While these approaches can prevent divergence, they can also exhibit different limitations e.g. resulting in too-sparse synaptic configurations or leading to poor scalability with increasing network size. A recently introduced meta-plasticity mechanism, termed annealed linear learning (ALL), dynamically reduces the learning rate as neuronal output increases, thereby preserving stable and interpretable fixed-point behavior of the output. However, the original formulation leads to an irreversible decay of the learning rate, preventing adaptation to changing environmental conditions. To address this, in the present study, we balance learning rate reduction at large outputs with recovery at small outputs and in addition introduce forgetting that gradually reduces synaptic weights. These extensions allow the system to discard outdated representations and adapt to novel input conditions. Analytical investigations demonstrate that the favorable output fixed-point properties of the original ALL framework are preserved under the extended rule. Furthermore, simulations with an artificial agent show that the proposed mechanism enables robust and fast re-learning and adaptation in changing environments.

The dentate gyrus (DG) decorrelates entorhinal inputs (pattern separation); area CA3 completes partial cues via recurrent autoassociation. The density of CA3 recurrent connectivity is contested, with estimates from [~]0.9% (Guzman et al., 2016) to [~]9-11% (Sammons et al., 2024). We ask how completion depends on recurrent connectivity (CRC) and whether the answer is intrinsic to CA3 dynamics or inherited from the DG front-end. Using a trisynaptic model that crosses two DG implementations (a point-LIF network with Santhakumar et al., 2005 topology; an abstract fixed-in-degree spiking network validated size-invariant to N = 107) with two CA3 autoassociators (binary k-WTA; spiking excitatory/inhibitory attractor) via burst-gated mossy-fiber detonators, we find: (i) completion in the binary CA3 improves monotonically with CRC and is robust across DG implementation; (ii) the spiking CA3 exhibits a runaway transition whose boundary is set by the product (active fraction x CRC), is not rescued by stronger feedback inhibition (8x), is insensitive to input overlap, and is size-invariant (N = 104-105); (iii) the two CA3 types have opposite failure modes (binary under-completes at low CRC; spiking runs away at high active-fraction x CRC) and a capacity/stability trade-off. Adult neurogenesis flips sign by the same logic: excitability-only young cells densify the code and collapse the spiking attractor, but if they recruit feedback inhibition they instead sparsen it and preserve recall. Consistent with classical sparse-coding attractor theory (Tsodyks and Feigelman, 1988), we propose that the contested CA3 connectivity is better read as an implementation-mode trade-off, and that the empirically sparse activity of CA3 (a {approx} 0.02-0.05) is the condition that lets a highly recurrent network perform stable autoassociation. Significance StatementHow densely CA3 pyramidal neurons interconnect is contested, with functional and anatomical estimates differing roughly tenfold. In a dentate-gyrus[->]CA3 model run across two DG and two CA3 implementations, we show this need not be a contradiction: whether higher recurrent connectivity helps or harms pattern completion depends on the CA3 computational mode and, above all, on how sparse CA3 activity is. A spiking attractor collapses once the product of active fraction and recurrent in-degree exceeds an approximately size-invariant threshold, whereas a hard-sparsity network is immune. Whether neurogenesis helps or harms depends on whether young neurons recruit inhibition: without it they destabilise an E/I CA3; with it they protect it. Sparse coding is thus the control variable for stable memory.

Neurons in primate visual cortical area V4 display tuning for multiple visual features, including color, shape, texture, and depth. Whether and how these neurons are organized into functional architectures remains largely unknown. Using two-photon calcium imaging in anesthetized macaques, we recorded responses of hundreds of V4 neurons to natural images and used these data to train deep convolutional neural network models (digital twins), obtaining synthetic images (SI) that maximized neuronal responses. Based on their SIs, these neurons clustered into classes that spatially matched the orientation, color, and curvature maps from intrinsic signal imaging. Furthermore, lesion study in digital twins revealed different integration rules for different neuron classes. Thus, digital twins of V4 neurons can be applied as a promising tool to characterize neurons fundamental features, which underlie the functional clustering in this area. HighlightsO_LINeurons in V4 are examined with two-photon calcium imaging and DCNN modeling (digital twins) C_LIO_LISynthetic images derived from neurons digital twins reveal distinct neuron groups C_LIO_LIThese groups match functional types defined by intrinsic signal optical imaging C_LIO_LILesion study in digital twins demonstrates neural mechanisms underlying feature tuning C_LI

The human visual system integrates both static and dynamic information to support form and shape perception, yet the computational principles underlying the integration of motion for object recognition remain unclear. Artificial neural networks (ANNs) offer a computational framework for developing and testing hypotheses about these principles: if ANNs trained on motion-related tasks develop representations that align with brain activity and support object categorization, this would suggest that the training objectives and architectural constraints of these networks may capture key aspects of motion processing in biological visual systems in general, and motion processing for object recognition, in particular. Here, we investigated this question using "object kinematograms", stimuli in which object form is conveyed solely through motion cues. We measured neural responses of two higher regions of the lateral and the dorsal visual pathways, respectively, with strong sensitivity to dynamic cues from objects: lateral occipitotemporal cortex (LOTbio), and left supramarginal gyrus (SMGlh), as well as primary visual cortex (V1). We compared brain responses to representations extracted from two neural networks: SlowFast, a dual-pathway architecture trained on action recognition that processes slow- and fast-varying visual information with cross-pathway integration, and DorsalNet, a model of the primate dorsal visual pathway trained on embodied self-motion estimation. Representational similarity analysis revealed distinct representational profiles across brain areas, demonstrating functional specialization in motion-based form processing. LOTbio was best characterized by the slow pathway of the SlowFast model, whereas SMGlh showed strong similarity to both models. Critically, we found that representations aligned with brain activity also better supported behavioral function: the full SlowFast model, incorporating both slow and fast pathways, outperformed other models in few-shot categorization of object kinematograms and showed the highest similarity to human perceptual judgments. These findings demonstrate that with appropriate inductive biases, specifically, dual-pathway architectures for multi-scale motion processing and training objectives focused on dynamic visual tasks, ANNs can develop functionally useful representations of motion-defined forms that exhibit better alignment with the visual regions involved in processing dynamic visual signals.