Search

While generative AI is now widespread and useful in society, there are potential risks of misuse, e.g., unconsciously influencing cognitive processes or decision-making. Although this causes a security problem in the cognitive domain, there has been no research about neural and computational mechanisms counteracting the impact of malicious generative AI in humans. We propose DecNefGAN, a novel framework that combines a generative adversarial system and a neural reinforcement model. More specifically, DecNefGAN bridges human and generative AI in a closed-loop system, with the AI creating stimuli that induce specific mental states, thus exerting external control over neural activity. The objective of the human is the opposite, to compete and reach an orthogonal mental state. This framework can contribute to elucidating how the human brain responds to and counteracts the potential influence of generative AI.

In human neuroscience, machine learning can help reveal lower-dimensional neural representations relevant to subjects' behavior. However, state-of-the-art models typically require large datasets to train, so are prone to overfitting on human neuroimaging data that often possess few samples but many input dimensions. Here, we capitalized on the fact that the features we seek in human neuroscience are precisely those relevant to subjects' behavior. We thus developed a Task-Relevant Autoencoder via Classifier Enhancement (TRACE), and tested its ability to extract behaviorally-relevant, separable representations compared to a standard autoencoder, a variational autoencoder, and principal component analysis for two severely truncated machine learning datasets. We then evaluated all models on fMRI data from 59 subjects who observed animals and objects. TRACE outperformed all models nearly unilaterally, showing up to 12% increased classification accuracy and up to 56% improvement in discovering "cleaner", task-relevant representations. These results showcase TRACE's potential for a wide variety of data related to human behavior., Comment: 41 pages, 11 figures, 5 tables including supplemental material

Collaboration frequently yields better results in decision making, learning, and haptic interactions than when these actions are performed individually. However, is collaboration always superior to solo actions, or do its benefits depend on whether collaborating individuals have different or the same roles? To answer this question, we asked human subjects to perform virtual-reality collaborative and individual beam transportation tasks. These tasks were simulated in real-time by coupling the motion of a pair of hand-held robotic manipulanda to the virtual beam using virtual spring-dampers. For the task to be considered successful, participants had to complete it within temporal and spatial constraints. While the visual feedback remained the same, the underlying dynamics of the beam were altered to create two distinctive task contexts which were determined by a moving pivot constraint. When the pivot was placed at the center of the beam, two hands contribute to the task with symmetric mechanical leverage (symmetric context). When the pivot was placed at the left side of the beam, two hands contribute to the task with asymmetric mechanical leverage (asymmetric context). Participants performed these task contexts either individually with both hands (solo), or collaboratively by pairing one hand with another one (dyads). We found that dyads in the asymmetric context performed better than solos. In contrast, solos performed the symmetric context better than dyads. Importantly, we found that two hands took different roles in the asymmetric context for both solos and dyads. In contrast, the contribution from each hand was statistically indistinguishable in the symmetric context. Our findings suggest that better performance in dyads than solos is not a general phenomenon, but rather that collaboration yields better performance only when role specialization emerges in dyadic interactions.

In several papers published in Biological Cybernetics in the 1980s and 1990s, Kawato and colleagues proposed computational models explaining how internal models are acquired in the cerebellum. These models were later supported by neurophysiological experiments using monkeys and neuroimaging experiments involving humans. These early studies influenced neuroscience from basic, sensory-motor control to higher cognitive functions. One of the most perplexing enigmas related to internal models is to understand the neural mechanisms that enable animals to learn large-dimensional problems with so few trials. Consciousness and metacognition -- the ability to monitor one's own thoughts, may be part of the solution to this enigma. Based on literature reviews of the past 20 years, here we propose a computational neuroscience model of metacognition. The model comprises a modular hierarchical reinforcement-learning architecture of parallel and layered, generative-inverse model pairs. In the prefrontal cortex, a distributed executive network called the "cognitive reality monitoring network" (CRMN) orchestrates conscious involvement of generative-inverse model pairs in perception and action. Based on mismatches between computations by generative and inverse models, as well as reward prediction errors, CRMN computes a "responsibility signal" that gates selection and learning of pairs in perception, action, and reinforcement learning. A high responsibility signal is given to the pairs that best capture the external world, that are competent in movements (small mismatch), and that are capable of reinforcement learning (small reward prediction error). CRMN selects pairs with higher responsibility signals as objects of metacognition, and consciousness is determined by the entropy of responsibility signals across all pairs., Comment: 23 pages, 3 figures. Fund number (ATLA) revised on Dec. 17

Fifty years have passed since David Marr, Masao Ito, and James Albus proposed seminal models of cerebellar functions. These models share the essential concept that parallel-fiber-Purkinje-cell synapses undergo plastic changes, guided by climbing-fiber activities during sensorimotor learning. However, they differ in several important respects, including holistic versus complementary roles of the cerebellum, pattern recognition versus control as computational objectives, potentiation versus depression of synaptic plasticity, teaching signals versus error signals transmitted by climbing-fibers, sparse expansion coding by granule cells, and cerebellar internal models. In this review, we evaluate the different features of the three models based on recent computational and experimental studies. While acknowledging that the three models have greatly advanced our understanding of cerebellar control mechanisms in eye movements and classical conditioning, we propose a new direction for computational frameworks of the cerebellum. That is, hierarchical reinforcement learning with multiple internal models., Comment: P.4, 5, 8, 10, 14, 16, 18, 22, and some references added

Artificial intelligence algorithms are capable of fantastic exploits, yet they are still grossly inefficient compared with the brain's ability to learn from few exemplars or solve problems that have not been explicitly defined. What is the secret that the evolution of human intelligence has unlocked? Generalization is one answer, but there is more to it. The brain does not directly solve difficult problems, it is able to recast them into new and more tractable problems. Here we propose a model whereby higher cognitive functions profoundly interact with reinforcement learning to drastically reduce the degrees of freedom of the search space, simplifying complex problems and fostering more efficient learning.

Psychiatric research has been hampered by an explanatory gap between psychiatric symptoms and their neural underpinnings, which has resulted in poor treatment outcomes. This situation has prompted us to shift from symptom-based diagnosis to data-driven diagnosis, aiming to redefine psychiatric disorders as disorders of neural circuitry. Promising candidates for data-driven diagnosis include resting-state functional connectivity MRI (rs-fcMRI)-based biomarkers. Although biomarkers have been developed with the aim of diagnosing patients and predicting the efficacy of therapy, the focus has shifted to the identification of biomarkers that represent therapeutic targets, which would allow for more personalized treatment approaches. This type of biomarker (i.e., theranostic biomarker) is expected to elucidate the disease mechanism of psychiatric conditions and to offer an individualized neural circuit-based therapeutic target based on the neural cause of a condition. To this end, researchers have developed rs-fcMRI-based biomarkers and investigated a causal relationship between potential biomarkers and disease-specific behavior using functional MRI (fMRI)-based neurofeedback on functional connectivity. In this review, we introduce recent approach for creating a theranostic biomarker, which consists mainly of two parts: (i) developing an rs-fcMRI-based biomarker that can predict diagnosis and/or symptoms with high accuracy, and (ii) the introduction of a proof-of-concept study investigating the relationship between normalizing the biomarker and symptom changes using fMRI-based neurofeedback. In parallel with the introduction of recent studies, we review rs-fcMRI-based biomarker and fMRI-based neurofeedback, focusing on the technological improvements and limitations associated with clinical use., Comment: 46 pages, 5 figures

By focusing on melancholic features with biological homogeneity, this study aimed to identify a small number of critical functional connections (FCs) that were specific only to the melancholic type of MDD. On the resting-state fMRI data, classifiers were developed to differentiate MDD patients from healthy controls (HCs). The classification accuracy was improved from 50 % (93 MDD and 93 HCs) to 70% (66 melancholic MDD and 66 HCs), when we specifically focused on the melancholic MDD with moderate or severer level of depressive symptoms. It showed 65% accuracy for the independent validation cohort. The biomarker score distribution showed improvements with escitalopram treatments, and also showed significant correlations with depression symptom scores. This classifier was specific to melancholic MDD, and it did not generalize in other mental disorders including autism spectrum disorder (ASD, 54% accuracy) and schizophrenia spectrum disorder (SSD, 45% accuracy). Among the identified 12 FCs from 9,316 FCs between whole brain anatomical node pairs, the left DLPFC / IFG region, which has most commonly been targeted for depression treatments, and its functional connections between Precuneus / PCC, and between right DLPFC / SMA areas had the highest contributions. Given the heterogeneity of the MDD, focusing on the melancholic features is the key to achieve high classification accuracy. The identified FCs specifically predicted the melancholic MDD and associated with subjective depressive symptoms. These results suggested key FCs of melancholic depression, and open doors to novel treatments targeting these regions in the future.

Obsessive-compulsive disorder (OCD) is a common psychiatric disorder with a lifetime prevalence of 2-3 percent. Recently, brain activity in the resting state is gathering attention as a new means of exploring altered functional connectivity in psychiatric disorders. Although previous resting-state functional magnetic resonance imaging studies investigated neurobiological abnormalities of patients with OCD, there are concerns that should be addressed. One concern is the validity of the hypothesis employed. Most studies used seed-based analysis of the fronto-striatal circuit, despite the potential for abnormalities in other regions. A hypothesis-free study is a promising approach in such a case, while it requires researchers to handle a dataset with large dimensions. Another concern is the reliability of biomarkers derived from a single dataset, which may be influenced by cohort-specific features. Here, by employing a recently developed machine-learning algorithm to avoid these concerns, we identified the first OCD biomarker that is generalized to an external dataset. We also demonstrated that the functional connectivities that contributed to the classification were widely distributed rather than locally constrained. Our generalizable classifier has the potential not only to deepen our understanding of the abnormal neural substrates of OCD but also to find use in clinical applications., Comment: 47 pages, 3 figures

Huang (arXiv:1612.03270) argues that the perceptual learning induced by our decoded neurofeedback method (DecNef) can be explained by Hebbian synaptic plasticity of connections between V1/V2 and V3/V4 rather than that within V1/V2, and that reinforcement learning at a cellular level should not be possible, and thus challenges our conclusions. In this reply, first, we show that Huang's model is not compatible with our data which strongly suggest plasticity within V1/V2. Second, the results of our new analysis show that spontaneous activity is not random but largely accounted for by low-dimensional components. These and other results indicate that learning to induce a stimulus-associated template voxel pattern in V1/V2 within as few as 100 trials is likely to be accomplished by hyper-resolution and sub-voxel level control by DecNef., Comment: 13 pages, 1 figure

Studies using real-time functional magnetic resonance imaging (rt-fMRI) have recently incorporated the decoding approach, allowing for fMRI to be used as a tool for manipulation of fine-grained neural activity. Because of the tremendous potential for clinical applications, certain questions regarding decoded neurofeedback (DecNef) must be addressed. Neurofeedback effects can last for months, but the short- to mid-term dynamics are not known. Specifically, can the same subjects learn to induce neural patterns in two opposite directions in different sessions? This leads to a further question, whether learning to reverse a neural pattern may be less effective after training to induce it in a previous session. Here we employed a within-subjects' design, with subjects undergoing DecNef training sequentially in opposite directions (up or down regulation of confidence judgements in a perceptual task), with the order counterbalanced across subjects. Behavioral results indicated that the manipulation was strongly influenced by the order and direction of neurofeedback. We therefore applied nonlinear mathematical modeling to parametrize four main consequences of DecNef: main effect of change in behavior, strength of down-regulation effect relative to up-regulation, maintenance of learning over sessions, and anterograde learning interference. Modeling results revealed that DecNef successfully induced bidirectional behavioral changes in different sessions. Furthermore, up-regulation was more sizable, and the effect was largely preserved even after an interval of one-week. Lastly, the second week effect was diminished as compared to the first week effect, indicating strong anterograde learning interference. These results suggest reinforcement learning characteristics of DecNef, and provide important constraints on its application to basic neuroscience, occupational and sports trainings, and therapies., Comment: 31 pages, 6 figures, 1 table

In human studies, how averaged activation in a brain region relates to human behavior has been extensively investigated. This approach has led to the finding that positive and negative facial preferences are represented by different brain regions. However, using a multi-voxel pattern induction method we found that different patterns of neural activations within the cingulate cortex (CC) play roles in representing opposite emotional states. In the present study, while neutrally-preferred faces were presented, activation patterns in the CC that corresponded to higher (or lower) preference were repeatedly induced by the pattern induction method. As a result, previously neutrally-preferred faces became more (or less) preferred. We conclude that a different activation pattern in the CC, rather than averaged activation in a different area, represents and causally determines positive or negative facial preference. This new approach may reveal importance in an activation pattern within a brain region in many cognitive functions., Comment: 49 pages, 7 figures