Biology subjects

Honnuraiah, S.

Publications and source records attributed to Honnuraiah, S..

7 recordsLinked to original sources

bioRxiv2025

SNNs Are Not Transformers (Yet): The Architectural Problems for SNNs in Modeling Long-Range Dependencies

Spiking neural networks (SNNs) have attracted growing interest for their ability to operate efficiently on low-power neuromorphic hardware, offering a biologically grounded route toward energy-efficient computation. However, despite advances in large-scale neuromorphic systems capable of simulating millions of spiking neurons and synapses, SNNs continue to underperform state-of-the-art (SOTA) artificial neural networks (ANNs) on complex sequence-processing tasks. Here, we present an explicit covering-number bound analysis for SNNs based on the non-leaky integrate and fire (nLIF) model. Leveraging recent work on causal partitions and local Lipschitz continuity, we derive a global Lipschitz constant and show that the sample complexity of nLIF networks scales quadratically with input sequence length. We analytically compare these bounds with those of Transformer and recurrent neural network (RNN) architectures, revealing fundamental constraints on how current SNNs process long-range dependencies. Finally, we show that these theoretical assumptions align with known cortical mechanisms, particularly inhibitory normalization and refractoriness, and discuss their implications for developing future neuromorphic architectures that more closely approximate biological computation.

neuroscience↗

bioRxiv2024

Biologically Realistic Computational Primitives of Neocortex Implemented on Neuromorphic Hardware Improve Vision Transformer Performance

Understanding the computational principles of the brain and replicating them on neuromorphic hardware and modern deep learning architectures is crucial for advancing neuro-inspired AI (NeuroAI). Here, we develop an experimentally-constrained biophysical network model of neocortical circuit motifs, focusing on layers 2-3 of the primary visual cortex (V1). We investigate the role of four major cortical interneuron classes in a competitive-cooperative computational primitive and validate these circuit motifs implemented soft winner-take-all (sWTA) computation for gain modulation, signal restoration, and context-dependent multistability. Using a novel parameter mapping technique, we configured IBMs TrueNorth (TN) chip to implement sWTA computations, mirroring biological neural dynamics. Retrospectively, we observed a strong correspondence between the biophysical model and the TN hardware parameters, particularly in the roles of four key inhibitory neuron classes: Parvalbumin (feedforward inhibition), Somatostatin (feedback inhibition), VIP (disinhibition), and LAMP5 (gain normalization). Moreover, the sparse coupling of this sWTA motif was also able to simulate a two-state neural state machine on the TN chip, replicating working memory dynamics essential for cognitive tasks. Additionally, integrating the sWTA computation as a preprocessing layer in the Vision Transformer (ViT) enhanced its performance on the MNIST digit classification task, demonstrating improved generalization to previously unseen data and suggesting a mechanism akin to zero-shot learning. Our approach provides a framework for translating brain-inspired computations to neuromorphic hardware, with potential applications on platforms like Intels Loihi2 and IBMs Northpole. By integrating biophysically accurate models with neuromorphic hardware and advanced machine learning techniques, we offer a comprehensive roadmap for embedding neural computation into NeuroAI systems.

neuroscience↗

bioRxiv2024

Cellular and circuit mechanisms underlying binocular vision

How binocular visual information is combined at the level of single neurons in the brain is not fully understood. Here, we show in mice that callosal input from the opposite visual cortex (V1) plays a critical role in this process. In vivo we find this callosal projection carries ipsilateral eye information and synapses exclusively onto binocular neurons. Using the presence of callosal input to identify binocular neurons in vitro, at the cellular level we show that binocular neurons are less excitable than monocular neurons due to high expression of Kv1 potassium channels. At the circuit level we find that only monocular neurons send callosal projections to the opposite V1, whereas binocular neurons do not. Finally, using dual-colour optogenetics we show that most binocular and monocular neurons receive direct input from the thalamus. In summary, we describe distinct cellular and circuit mechanisms underlying processing of binocular visual information in mouse V1.

neuroscience↗

bioRxiv2024

An Anatomical and Physiological Basis for Coincidence Detection Across Time Scales in the Auditory System

Animals navigate the auditory world by recognizing complex sounds, from the rustle of a predator to the call of a potential mate. This ability depends in part on the octopus cells of the auditory brainstem, which respond to multiple frequencies that change over time, as occurs in natural stimuli. Unlike the average neuron, which integrates inputs over time on the order of tens of milliseconds, octopus cells must detect momentary coincidence of excitatory inputs from the cochlea during an ongoing sound on both the millisecond and submillisecond time scale. Here, we show that octopus cells receive inhibitory inputs on their dendrites that enhance opportunities for coincidence detection in the cell body, thereby allowing for responses both to rapid onsets at the beginning of a sound and to frequency modulations during the sound. This mechanism is crucial for the fundamental process of integrating the synchronized frequencies of natural auditory signals over time.

neuroscience↗

bioRxiv2021

Higher synaptic threshold for NMDA spike generation in human neurons

Neurons receive synaptic input primarily onto their dendrites. While we know much about the electrical properties of dendrites in rodents, we have only just started to describe their properties in the human brain. Here we investigate the capacity of human neurons to generate NMDA spikes. We find that dendritic iontophoresis of glutamate, as well as local dendritic synaptic stimulation, can evoke NMDA spikes in dendrites of human layer 2/3 pyramidal neurons. Surprisingly, however, the capacity to evoke NMDA spikes in human neurons was significantly reduced compared to that in rodent neurons. Simulations in morphologically realistic models indicated that human neurons have a higher synaptic threshold for NMDA spike generation. Using a simplified model, we show that this is primarily due to the larger diameter of human dendrites. In summary, we find reduced NMDA spike generation in human compared to rodent neurons due to the larger diameter of human dendrites.

neuroscience↗

bioRxiv2019

Superior colliculus modulates cortical coding of somatosensory information

The cortex sends a direct projection to the superior colliculus. What is largely unknown is whether (and if so how) the superior colliculus modulates activity in the cortex. Here, we directly investigate this issue, showing that optogenetic activation of superior colliculus changes the input-output relationship of neurons in somatosensory cortex during whisker movement, enhancing responses to low amplitude whisker deflections. While there is no direct pathway from superior colliculus to somatosensory cortex, we found that activation of superior colliculus drives spiking in the posterior medial (POm) nucleus of the thalamus via a powerful monosynaptic pathway. Furthermore, POm neurons receiving input from superior colliculus provide excitatory input to somatosensory cortex. Silencing POm abolished the capacity of superior colliculus to modulate cortical whisker responses. Our findings indicate that the superior colliculus, which plays a key role in attention, modulates sensory processing in somatosensory cortex via a powerful disynaptic pathway through the thalamus.

neuroscience↗

bioRxiv2017

Conditioning by Subthreshold Synaptic Input Changes the Characteristic Firing Pattern of CA3 Hippocampal Neurons

Unlike synaptic strength, intrinsic excitability is assumed to be a stable property of neurons. For example, learning of somatic conductances is generally not incorporated into computational models, and the discharge pattern of neurons in response to test stimuli is frequently used as a basis for phenotypic classification. However, it is increasingly evident that signal processing properties of neurons are more generally plastic on the timescale of minutes. Here we demonstrate that the intrinsic firing patterns of CA3 neurons of the rat hippocampus in vitro undergo rapid long-term plasticity in response to a few minutes of only subthreshold synaptic conditioning. This plasticity on the spike-timing could also be induced by intrasomatic injection of subthreshold depolarizing pulses and was blocked by kinase inhibitors, indicating that discharge dynamics are modulated locally. Cluster analysis of firing patterns before and after conditioning revealed systematic transitions towards adapting and intrinsic burst behaviours, irrespective of the patterns initially exhibited by the cells. We used a conductance-based model to decide appropriate pharmacological blockade, and found that the observed transitions are likely due to recruitment of calcium and M-type potassium conductances. We conclude that CA3 neurons adapt their conductance profile to the subthreshold activity of their input, so that their intrinsic firing pattern is not a static signature, but rather a reflection of their history of subthreshold activity. In this way, recurrent output from CA3 neurons may collectively shape the temporal dynamics of their embedding circuits.\n\nNew & NoteworthyDespite being widely conserved across the animal phyla, it is still a mystery why nerve cells present diverse discharge dynamics upon somatic step currents. Adding a new timing dimension to the intrinsic plasticity literature, here we show that CA3 neurons rapidly adapt through the space of known firing patterns in response to the subthreshold signals that they receive from their embedding circuit. This result implies that CA3 neurons collectively adjust their network processing to the temporal statistics of their circuit.

neuroscience↗