Behavioral tasks can be solved employing various strategies. Sometimes, different strategies result in the same observable behavior, making them latent. In this study, we infer the latent behavioral strategy used by monkeys in a working memory updating task by comparing the representational geometry of two prefrontal regions - the lateral prefrontal cortex (LPFC) and the prearcuate cortex (PAC) - with that of recurrent neural network (RNN) models trained to solve the task using different strategies. We found that neural activity patterns in both LPFC and PAC align with only one of the proposed strategies, suggesting that monkeys employ this latent strategy to perform the task. These findings open avenues for investigating the processes that lead to strategy learning and the decision-making mechanisms that determine which strategies are chosen when multiple options are available.

Paralysis affects many people worldwide, and the people affected often suffer from impaired communication. We developed a microelectrode-based Brain-Computer Interface (BCI) for enabling communication in patients affected by paralysis, and implanted it in a patient with Multiple System Atrophy (MSA), a neurodegenerative disease that causes widespread neural symptoms including paralysis. To verify the effectiveness of the BCI system, it was also tested by implanting it in a non-human primate (NHP). Data from the human and NHP were used to train binary classifiers two different types of machine learning models: a Linear Discriminant Analysis (LDA) model, and a Long Short-Term Memory (LSTM)-based Artificial Neural Network (ANN). The LDA model performed at up to 72.7 % accuracy for binary decoding in the human patient, however, performance was highly variable and was much lower on most recording days. The BCI system was able to accurately decode movement vs non-movement in the NHP (accuracy using LDA: 82.7 ± 3.3 %, LSTM: 83.7 ± 2.2 %, 95 % confidence intervals), however it was not able to with recordings from the human patient (accuracy using LDA: 47.0 ± 5.1 %, LSTM: 44.6 ± 9.9 %, 95 % confidence intervals). We discuss how neurodegenerative diseases such as MSA can impede BCI-based communication, and postulate on the mechanisms by which this may occur.

Working memory updating is an important executive process. Here, we study the single-neuron mechanisms involved in updating versus protecting memory from distractors in the macaque prefrontal cortex. We recorded single-neuron activity from the lateral prefrontal cortex (LPFC) and prearcuate cortex (PAC) while male monkeys performed a task that required them to update their memory of target locations while ignoring distractors. Our findings revealed that neurons in the PAC signaled updated memory locations ∼100 ms after stimulus onset, significantly faster than the ∼400 ms observed in the LPFC. Additionally, PAC neurons exhibited longer encoding of distractor information. Population decoding analyses further indicated that distractor information was maintained in orthogonal subspaces from target information in both regions, minimizing interference. These results demonstrate the distinct temporal dynamics in memory updating processes between the PAC and LPFC and highlight the interplay between robust memory maintenance and updating, suggesting that local neural mechanisms may contribute to these processes.

The functional properties of a network depend on its connectivity, which includes the strength of its inputs and the strength of the connections between its units, or recurrent connectivity. Because we lack a detailed description of the recurrent connectivity in the lateral prefrontal cortex of primates, we developed an indirect method to estimate it. This method leverages the elevated noise correlation of mutually-connected units. To estimate the connectivity of prefrontal regions, we trained recurrent neural network models with varying percentages of bump attractor connectivity and noise levels to match the noise correlation properties observed in two specific prefrontal regions: the dorsolateral prefrontal cortex and the frontal eye field. We found that models initialized with approximately 20% and 7.5% bump attractor connectivity closely matched the noise correlation properties of the frontal eye field and dorsolateral prefrontal cortex, respectively. These findings suggest that the different percentages of bump attractor connectivity may reflect distinct functional roles of these brain regions. Specifically, lower percentages of bump attractor units, associated with higher-dimensional representations, likely support more abstract neural representations in more anterior regions.

This paper presents an extension of the "neural replay thought experiment" as a heuristic tool for probing the reader’s implicit assumptions regarding the relationship between physical and mental phenomena. By systematically varying the physical instantiation of a recorded neural activity pattern, we aim to elicit divergent intuitions about the persistence of consciousness and intelligence. The analysis focuses on possible arguments asserting the disappearance or alteration of consciousness and intelligence across these scenarios, alongside corresponding counter-arguments. While not exhaustive, this exploration highlights the challenges in articulating a robust physical basis for mental phenomena and underscores the counterintuitive consequences that often arise from seemingly straightforward assumptions. The discussion briefly considers the broader implications of this analysis for theories of mind and intelligence.

General Description. This dataset consists of data from four BMI experiments performed on two adult macaques. Three of the experiments were joystick controlled tasks, and one of them was center-out reaching task. The macaques were able to use a wireless integrated system to control a robotic platform, over which they were sitting, to achieve independent mobility using the neuronal activity in their motor cortices. The activity of populations of single neurons was recorded using multiple electrode arrays implanted in the arm region of primary motor cortex. A general description is provided below: A titanium head post (Crist Instruments, MD, USA) was affixed prior to implantation of microelectrode arrays in both NHPs. In NHP-A, 4 microelectrode arrays containing 16 electrodes each, and in NHP-B, 1 microelectrode array containing 100 electrodes were implanted in the hand/arm region of the left primary motor cortex respectively. Spike signals were acquired using an in-house 100-channel wireless neural recording system, which is sampled at 13 KHz. The wide-band signals were then band-pass filtered between 300 to 3000 Hz to remove low-frequency components. The threshold for spike deterction was found using the formula: (Thr = 5σ; σ = median(|x|/0.6745), where x is the filtered signal, and σ is an estimate of the standard deviation of the background noise. The behavorial task was to make a robotic wheelchair bound control its motion through a three-directional spring-loaded joystick (Experiment 1, 2, and 3). The experiment comprised of four tasks - a) turning 90° right, b) moving forward by 2m, c) turning 90° left, and d) staying still for 5 seconds (stop task). Successful task completion varied from experiment to experiment. Experiment 4 also involved joystick control but the primate was trained to perform classical center-out task. Data for Experiment 1 and 3 are already publicly available at: https://osf.io/dce96/. However, a detailed description is also provided here. The data are grouped in form of folders containing data for NHP-1/2-Set 1/2. For the folder, NHP 1 Set 1, experiment 1 data consists of sessions 1,2,3; expt 3: 5,6,7,8. Similarly for Set 2: expt-1: 3,4,5,10,11; expt 3: 8,9. For the folder, NHP 2 Set 1, expt 1 consists of sessions 10,11,12,13,18,19,20,21; expt 3: 15,16,17,24. For the folder. Similarly for Set 2: expt-1: 1,2,3,10,11,12,13; expt 3: 6,7,8,9. Possible use cases. These data are ideal for designing, training, and testing iBMI decoders. We expect that the dataset will be valuable for researchers who wish to design improved models of sensorimotor cortical spiking or provide an equal footing for comparing different iBMI decoders. We also hope to inspire more work along neuromorphic lines and use of online Reinforcement Learning algorithms for iBMI decoders. Variable names. Each file from Experiment 1 and 3 contains data in the following format. 1. joystick_adfreq: The frequency of operation of the joystick. 2. X_Voltage: The voltage reading corresponding to the x-coordinate (while joystick operation). 3. Y_Voltage: The voltage reading corresponding to the y-coordinate (while joystick operation). 4. Spike_data(Channel Number): The Channel Number corresponding to which the neuronal data is recorded. 5. Spike_data(Cluster): Descripting the cluster on which the channels are placed. 6. Spike_data(Spike Times): The timestamp corresponding to the detection of a spike. 7. Spike_data(Spike Number): The total number of spikes calculated for a particular trial from a particular channel. 8. Spike_data(Mean Spike Waveform): The mean neuronal data (for that trial from a particular channel) describing a spike. 9. Spike_data(Spike Amplitude): The mean spike amplitude of that particular channel. 10. IMETrainingData(SentSignals): The truth labels corresponding to a particular trial. 11. IMETrainingData(Timestamps): Time stamps corresponding to each sent signal (data). 12. IMETrainingData(ReasonFail): String data; Reason if the trial failed. 13. IMETrainingData(TrialOutcomes): Trial results corresponding to successful or unsuccessful! 14. IMETrainingData(StartTime): corresponding to the beginning of each trial. 15. IMETrainingData(EndTime): corresponding to the end of each trial. For files in Experiment 2 and 4, 1. targetTest_Acc: Corresponding direction of the joystick recorded for each trial. (decoded using the decoder) 2. targetTrain: Ground truth label, corresponding to the actual direction of the joystick (for each trial) 3. testingSet_Acc: Number of spike counts from each channel (used for testing corresponding to all the sessions) 4. trainingSet: Number of spike counts from each channel (used for calibration, mostly) Contact Information. We would be delighted to hear from you if you find this dataset valuable, especially if it leads to publication. Corresponding author: A. Ghosh ; A. Basu . Citation. A. Ghosh, S. Shaikh, P. S. V. Sun, C. Libedinsky, R. So, N. Lin, H. Chen, Z. Wang, A. Basu, "Low-complexity Reinforcement Learning Decoders for Autonomous, Scalable, Neuromorphic intra-cortical Brain Machine Interfaces," IEEE Transaction on Neural Networks and Learning Systems (Under review)

High-density single-unit recording among closely spaced neurons over long durations is crucial for understanding the cellular-level functional architecture of the brain. Existing brain probes, however, either laterally sample neurons far more sparsely than neuronal density in the same cortical layer or suffer from long-term instability due to electrode modification, preventing precise neuron-to-neuron interrogation in local microcircuits. Here, we report a monolithic graphene-edge probe (NeuroEdge) that achieves single-unit recording at neuronal density (16 electrodes within 100 µm diameter). We fabricate NeuroEdge using self-assembled reduced graphene oxide nanoflakes to form an electrochemically active tip consisting of exposed graphene nanoedges and electrolyte-filling nanotunnels, achieving an ultralow specific impedance of 20 MΩ µm2. In vivo experiments over 5 months demonstrate recording at a high signal-to-noise ratio (>20 dB) and reliable interrogation of neighboring neurons. We also show that NeuroEdge can record from a single auditory cortical layer and reveal heterogeneities in the acoustic frequency response and dynamic connectivity among neighboring neurons. NeuroEdge provides a tool for precisely interrogating local microcircuits at the density of neurons in the brain. ### Competing Interest Statement Yunxia Jin, John S. Ho, Chwee Teck Lim and Yuxin Liu are on a patent application filed by the National University of Singapore relating to this work. All the other authors declare no competing interests. Advanced Research and Technology Innovation Centre, HFM-RP6 National University of Singapore National Research Foundation Singapore, NRFF2017-07 Ministry of Education Singapore, MOE2016-T3-1-004 National University of Singapore, A-8001301-00-00, A-0009363-04-00

Background. Brain–machine interfaces (BMIs) offer users the ability to directly communicate with digital devices through neural signals decoded with machine learning (ML)-based algorithms. Spiking Neural Networks (SNNs) are a type of Artificial Neural Network (ANN) that operate on neural spikes instead of continuous scalar outputs. Compared to traditional ANNs, SNNs perform fewer computations, use less memory, and mimic biological neurons better. However, SNNs only retain information for short durations, limiting their ability to capture long-term dependencies in time-variant data. Here, we propose a novel spike-weighted SNN with spiking long short-term memory (swSNN-SLSTM) for a regression problem. Spike-weighting captures neuronal firing rate instead of membrane potential, and the SLSTM layer captures long-term dependencies. Methods. We compared the performance of various ML algorithms during decoding directional movements, using a dataset of microelectrode recordings from a macaque during a directional joystick task, and also an open-source dataset. We thus quantified how swSNN-SLSTM performed compared to existing ML models: an unscented Kalman filter, LSTM-based ANN, and membrane-based SNN techniques. Result. The proposed swSNN-SLSTM outperforms both the unscented Kalman filter, the LSTM-based ANN, and the membrane based SNN technique. This shows that incorporating SLSTM can better capture long-term dependencies within neural data. Also, our proposed swSNN-SLSTM algorithm shows promise in reducing power consumption and lowering heat dissipation in implanted BMIs.

The lateral prefrontal cortex encodes working memory and motor preparation information, but the underlying neural mechanisms are debated. Recurrent neural network models relying on persistent neural activity have been challenged by the observation of periods of absent activity and information during memory maintenance, implying the existence of activity-silent mechanisms. To assess whether activity-silent mechanisms are needed for working memory maintenance, we recorded neural activity in macaque prefrontal cortex during a delayed-saccade task. We replicated the observation of periods of absent activity and decreased information between bursts of gamma power, but we show that these results are consistent with models that rely exclusively on persistent activity. However, an assessment of the length of periods with absent selective activity across the population revealed that activity-silent mechanisms are indeed needed to maintain memory information, although this is only evidenced in a small fraction of trials. ### Competing Interest Statement The authors have declared no competing interest.

ABSTRACTThe stochastic drift-diffusion model proposes that the variability in reaction time is due to randomness during the accumulation of evidence until a decision threshold is reached. However, the neural mechanisms that explain both the randomness and implementation of the decision threshold in the model remain unclear. Here we address these questions using the dynamical systems approach to analyze primate frontal eye field activity and using microstimulation for causal manipulations. We built a mechanistic model in which signals associated with motor plans are bumped out of their attractor state by go-cue signals that emerge ∼60 ms after the go cue. The network then travels through a transition subspace towards a movement-initation subspace that emerges ∼35 ms before movement onset and implements the decision threshold. We postulate that the randomness in evidence accumulation, and hence in reaction times, is explained by the amplification of noise during movement preparation by the geometry of the frontal eye field manifold.

The dorsolateral prefrontal cortex (dlPFC) is composed of multiple anatomically defined regions involved in higher-order cognitive processes, including working memory and selective attention. It is organized in an anterior–posterior global gradient where posterior regions track changes in the environment, whereas anterior regions support abstract neural representations. However, it remains unknown if such a global gradient results from a smooth gradient that spans regions or an emergent property arising from functionally distinct regions, that is, an areal gradient. Here, we recorded single neurons in the dlPFC of nonhuman primates trained to perform a memory-guided saccade task with an interfering distractor and analyzed their physiological properties along the anterior–posterior axis. We found that these physiological properties were best described by an areal gradient. Further, population analyses revealed that there is a distributed representation of spatial information across the dlPFC. Our results validate the functional boundaries between anatomically defined dlPFC regions and highlight the distributed nature of computations underlying working memory across the dlPFC. SIGNIFICANCE STATEMENT Activity of frontal lobe regions is known to possess an anterior–posterior functional gradient. However, it is not known whether this gradient is the result of individual brain regions organized in a gradient (like a staircase), or a smooth gradient that spans regions (like a slide). Analysis of physiological properties of individual neurons in the primate frontal regions suggest that individual regions are organized as a gradient, rather than a smooth gradient. At the population level, working memory was more prominent in posterior regions, although it was also present in anterior regions. This is consistent with the functional segregation of brain regions that is also observed in other systems (i.e., the visual system).

Since 2020, synchrotron radiation facilities in several Asia-Pacific countries have been collaborating in a major project called "SYNAPSE" (Synchrotrons for Neuroscience: an Asia-Pacific Scientific Enterprise). They use x-ray imaging to attack in a coordinated fashion one of the major issues in modern science: the structure of animal and human brains, including neurons and connections. The objective is to develop Google-like maps also including detailed structural and functional information for selected regions of interest. The sheer mass of data needed for the objective poses huge problems for the acquisition, processing, storage and use of images. In order to complete the task within a reasonable time, the key element of the SYNAPSE strategy is the parallel and coordinated work of several facilities on the same specimens. This article reviews different aspects of the enterprise, including the foundations of synchrotron radiation, coherence and of its role in advanced imaging, electron accelerators, x-ray optics and detectors. This will provide the foundation for an extensive presentation of the different components of SYNAPSE, with an overview of results already obtained within the consortium.

Single-neuron-level explanations have been the gold standard in neuroscience for decades. Recently, however, neural-network-level explanations have become increasingly popular. This increase in popularity is driven by the fact that the analysis of neural networks can solve problems that cannot be addressed by analyzing neurons independently. In this opinion article, I argue that while both frameworks employ the same general logic to link physical and mental phenomena, in many cases the neural network framework provides better explanatory objects to understand representations and computations related to mental phenomena. I discuss what constitutes a mechanistic explanation in neural systems, provide examples, and conclude by highlighting a number of the challenges and considerations associated with the use of analyses of neural networks to study brain function.

Implantable brain-machine interface (BMI) technology provides a promising solution for restoring independent locomotion for people with tetraplegia. However, current BMI systems used in clinical trials have not been widely adopted due to a number of shortcomings, one of which is the need for a wired connection. In this chapter, we present an example of a wireless BMI system implemented on a macaque model, showing that animals were able to use wirelessly transmitted neural signals for self-driving. From this example we discuss different aspects of design for such a BMI system, including data acquisition, signal processing, decoding/control algorithms, and neural responses when using the system. Future developments in this research field will enable BMI for locomotion control to become a reality.

Navigation to multiple cued reward locations has been increasingly used to study rodent learning. Though deep reinforcement learning agents have been shown to be able to learn the task, they are not biologically plausible. Biologically plausible classic actor-critic agents have been shown to learn to navigate to single reward locations, but which biologically plausible agents are able to learn multiple cue-reward location tasks has remained unclear. In this computational study, we show versions of classic agents that learn to navigate to a single reward location, and adapt to reward location displacement, but are not able to learn multiple paired association navigation. The limitation is overcome by an agent in which place cell and cue information are first processed by a feedforward nonlinear hidden layer with synapses to the actor and critic subject to temporal difference error-modulated plasticity. Faster learning is obtained when the feedforward layer is replaced by a recurrent reservoir network.

Objective: Peripheral neural interface (PNI) with a stable integration of synthetic elements with neural tissue is key for successfulneuro-prosthetic applications. An inevitable phenomenon of reactive fibrosis is a primary hurdle for long term functionality of PNIs. This proof-of-concept study aimed to fabricate and test a novel, stable PNI that harnesses fibro-axonal outgrowth at the nerve end and includes fibrosis in the design. Methods: Two non-human primates were implanted with Substrate-guided, Tissue-Electrode Encapsulation and Integration (STEER) PNIs. The implant included a 3D printed guide that strove to steer the regrowing nerve towards encapsulation of the electrodes into a fibro-axonal tissue. After four months from implantation, we performed electrophysiological measurements to test STEER's functionality and examined the macro and micro- morphology of the outgrowth tissue. Results: We observed a highly structured fibro-axonal composite within the STEER PNI. A conduction of intracranially generated action potentials was successfully recorded across the neural interface. Immunohistology demonstrated uniquely configured laminae of myelinated axons encasing the implant. Conclusion: STEER PNI reconfigured the structure of the fibro-axonal tissue and facilitated long-term functionality and stability of the neural interface. Significance: The results point to the feasibility of our concept for creating a stable PNI with long-term electrophysiologic functionality by using simple design and materials.

Implanted microelectrode arrays can directly pick up electrode signals from the primary motor cortex (M1) during movement, and brain-machine interfaces (BMIs) can decode these signals to predict the directions of contemporaneous movements. However, it is not well known how much each individual input is responsible for the overall performance of a BMI decoder. In this paper, we seek to quantify how much each channel contributes to an artificial neural network (ANN)-based decoder, by measuring how much the removal of each individual channel degrades the accuracy of the output. If information on movement direction was equally distributed among channels, then the removal of one would have a minimal effect on decoder accuracy. On the other hand, if that information was distributed sparsely, then the removal of specific information-rich channels would significantly lower decoder accuracy. We found that for most channels, their removal did not significantly affect decoder performance. However, for a subset of channels (16 out of 61), removing them significantly reduced the decoder accuracy. This suggests that information is not uniformly distributed among the recording channels. We propose examining these channels further to optimize BMIs more effectively, as well as understand how M1 functions at the neuronal level.

This paper is directed at scientists interested in the relationship between the physical and the mental. My goal is to provide an accessible platform to expose and analyze the readers’ (often implicit) assumptions about the relationship between physical and mental phenomena. To this end, I developed an extension of the “neural replay thought experiment”

Objective. With practice, the control of brain-computer interfaces (BCI) would improve over time; the neural correlate for such learning had not been well studied. We demonstrated here that monkeys controlling a motor BCI using a linear discriminant analysis (LDA) decoder could learn to make the firing patterns of the recorded neurons more distinct over a short period of time for different output classes to improve task performance. Approach. Using an LDA decoder, we studied two Macaque monkeys implanted with microelectrode arrays as they controlled the movement of a mobile robotic platform. The LDA decoder mapped high-dimensional neuronal firing patterns linearly onto a lower-dimensional linear discriminant (LD) space, and we studied the changes in the spatial coordinates of these neural signals in the LD space over time, and their correspondence to trial performance. Direction selectivity was quantified with permutation feature importance (FI). Main results. We observed that, within individual sessions, there was a tendency for the points in the LD space encoding different directions to diverge, leading to fewer misclassification errors, and, hence, improvement in task accuracy. Accuracy was correlated with the presence of channels with strong directional preference (i.e. high FI), as well as a varied population code (i.e. high variance in FI distribution). Significance. We emphasized the importance of studying the short-term/intra-sessional variations in neural representations during the use of BCI. Over the course of individual sessions, both monkeys could modulate their neural activities to create increasingly distinct neural representations.