Beyond Hebb: exclusive-OR and biological learning

Flexibly Photo-Regulated Brain-Inspired Functions in Flexible Neuromorphic Transistors

As an attractive prototype for neuromorphic computing, the difficultly attained three-terminal platforms have specific advantages in implementing the brain-inspired functions. Also, in these devices, the most utilized mechanisms are confined to the electrical gate-controlled ionic migrations, which are sensitive to the device defects and stoichiometric ratio. The resultant memristive responses have fluctuant characteristics, which have adverse influences on the neural emulations. Herein, we designed a specific transistor platform with light-regulated ambipolar memory characteristics. Also, based on its gentle processes of charge trapping, we obtain the impressive memristive performances featured by smooth responses and long-term endurable characteristics. The optoelectronic samples were also fabricated on flexible substrates successfully. Interestingly, based on the optoelectronic signals of the flexible devices, we endow the desirable optical processes with the brain-inspired emulations. We can flexibly emulate the light-inspired learning-memory functions in a synapse and further devise the advanced synapse array. More importantly, through this versatile platform, we investigate the mutual regulation of excitation and inhibition and implement their sensitive-mode transformations and the homeostasis property, which is conducive to ensuring the stability of overall neural activity. Furthermore, our flexible optoelectronic platform achieves high classification accuracy when implemented in artificial neural network simulations. This work demonstrates the advantages of the optoelectronic platform in implementing the significant brain-inspired functions and provides an insight into the future integration of visible sensing in flexible optoelectronic transistor platforms.

Oscillations in networks of networks stem from adaptive nodes with memory

We present an analytical framework that allows the quantitative study of statistical dynamic properties of networks with adaptive nodes that have memory and is used to examine the emergence of oscillations in networks with response failures. The frequency of the oscillations was quantitatively found to increase with the excitability of the nodes and with the average degree of the network and to decrease with delays between nodes. For networks of networks, diverse cluster oscillation modes were found as a function of the topology. Analytical results are in agreement with large-scale simulations and open the horizon for understanding network dynamics composed of finite memory nodes as well as their different phases of activity.

Long-Term Homeostatic Properties Complementary to Hebbian Rules in CuPc-Based Multifunctional Memristor

Most simulations of neuroplasticity in memristors, which are potentially used to develop artificial synapses, are confined to the basic biological Hebbian rules. However, the simplex rules potentially can induce excessive excitation/inhibition, even collapse of neural activities, because they neglect the properties of long-term homeostasis involved in the frameworks of realistic neural networks. Here, we develop organic CuPc-based memristors of which excitatory and inhibitory conductivities can implement both Hebbian rules and homeostatic plasticity, complementary to Hebbian patterns and conductive to the long-term homeostasis. In another adaptive situation for homeostasis, in thicker samples, the overall excitement under periodic moderate stimuli tends to decrease and be recovered under intense inputs. Interestingly, the prototypes can be equipped with bio-inspired habituation and sensitization functions outperforming the conventional simplified algorithms. They mutually regulate each other to obtain the homeostasis. Therefore, we develop a novel versatile memristor with advanced synaptic homeostasis for comprehensive neural functions.

Dual logic and cerebral coordinates for reciprocal interaction in eye contact

In order to scientifically study the human brain’s response to face-to-face social interaction, the scientific method itself needs to be reconsidered so that both quantitative observation and symbolic reasoning can be adapted to the situation where the observer is also observed. In light of the recent development of dyadic fMRI which can directly observe dyadic brain interacting in one MRI scanner, this paper aims to establish a new form of logic, dual logic, which provides a theoretical platform for deductive reasoning in a complementary dual system with emergence mechanism. Applying the dual logic in the dfMRI experimental design and data analysis, the exogenous and endogenous dual systems in the BOLD responses can be identified; the non-reciprocal responses in the dual system can be suppressed; a cerebral coordinate for reciprocal interaction can be generated. Elucidated by dual logic deductions, the cerebral coordinate for reciprocal interaction suggests: the exogenous and endogenous systems consist of the empathy network and the mentalization network respectively; the default-mode network emerges from the resting state to activation in the endogenous system during reciprocal interaction; the cingulate plays an essential role in the emergence from the exogenous system to the endogenous system. Overall, the dual logic deductions are supported by the dfMRI experimental results and are consistent with current literature. Both the theoretical framework and experimental method set the stage to formally apply the scientific method in studying complex social interaction.

Error correction and fast detectors implemented by ultrafast neuronal plasticity

We experimentally show that the neuron functions as a precise time integrator, where the accumulated changes in neuronal response latencies, under complex and random stimulation patterns, are solely a function of a global quantity, the average time lag between stimulations. In contrast, momentary leaps in the neuronal response latency follow trends of consecutive stimulations, indicating ultrafast neuronal plasticity. On a circuit level, this ultrafast neuronal plasticity phenomenon implements error-correction mechanisms and fast detectors for misplaced stimulations. Additionally, at moderate (high) stimulation rates this phenomenon destabilizes (stabilizes) a periodic neuronal activity disrupted by misplaced stimulations.

A reinforcement learning framework for spiking networks with dynamic synapses

An integration of both the Hebbian-based and reinforcement learning (RL) rules is presented for dynamic synapses. The proposed framework permits the Hebbian rule to update the hidden synaptic model parameters regulating the synaptic response rather than the synaptic weights. This is performed using both the value and the sign of the temporal difference in the reward signal after each trial. Applying this framework, a spiking network with spike-timing-dependent synapses is tested to learn the exclusive-OR computation on a temporally coded basis. Reward values are calculated with the distance between the output spike train of the network and a reference target one. Results show that the network is able to capture the required dynamics and that the proposed framework can reveal indeed an integrated version of Hebbian and RL. The proposed framework is tractable and less computationally expensive. The framework is applicable to a wide class of synaptic models and is not restricted to the used neural representation. This generality, along with the reported results, supports adopting the introduced approach to benefit from the biologically plausible synaptic models in a wide range of intuitive signal processing.

Network inference using asynchronously updated kinetic Ising model

Network structures are reconstructed from dynamical data by respectively naive mean field (nMF) and Thouless-Anderson-Palmer (TAP) approximations. TAP approximation adds simple corrections to the nMF approximation, taking into account the effect of the focused spin on itself via its influence on other neighboring spins. For TAP approximation, we use two methods to reconstruct the network: (a) iterative method; (b) casting the inference formula to a set of cubic equations and solving it directly. We investigate inference of the asymmetric Sherrington-Kirkpatrick (aS-K) model using asynchronous update. The solutions of the set of cubic equations depend on temperature T in the aS-K model, and a critical temperature T(c)≈2.1 is found. The two methods for TAP approximation produce the same results when the iterative method is convergent. Compared to nMF, TAP is somewhat better at low temperatures, but approaches the same performance as temperature increases. Both nMF and TAP approximation reconstruct better for longer data length L, but for the degree of improvement, TAP performs better than nMF.

Multidimensional fitting for multivariate data analysis

Large multidimensional data matrices are frequent in biology. However, statistical methods often have difficulties dealing with such matrices because they contain very complex data sets. Consequently variable selection and dimensionality reduction methods are often used to reduce matrix complexity, although at the expense of information conservation. A new method derived from multidimensional scaling (MDS) is presented for the case where two matrices are available to describe the same population. The presented method transforms one of the matrices, called the target matrix, with some constraints to make it fit with the second matrix, referred to as the reference matrix. The fitting to the reference matrix is performed on the distances computed for the two matrices, and the transformation depends on the problem at hand. A special feature of this method is that a variable can be only partially modified. The method is applied on the exclusive-or (XOR) problem and then on a biological application with large-scale gene expression data.

Learning as a phenomenon occurring in a critical state

Recent physiological measurements have provided clear evidence about scale-free avalanche brain activity and EEG spectra, feeding the classical enigma of how such a chaotic system can ever learn or respond in a controlled and reproducible way. Models for learning, like neural networks or perceptrons, have traditionally avoided strong fluctuations. Conversely, we propose that brain activity having features typical of systems at a critical point represents a crucial ingredient for learning. We present here a study that provides unique insights toward the understanding of the problem. Our model is able to reproduce quantitatively the experimentally observed critical state of the brain and, at the same time, learns and remembers logical rules including the exclusive OR, which has posed difficulties to several previous attempts. We implement the model on a network with topological properties close to the functionality network in real brains. Learning occurs via plastic adaptation of synaptic strengths and exhibits universal features. We find that the learning performance and the average time required to learn are controlled by the strength of plastic adaptation, in a way independent of the specific task assigned to the system. Even complex rules can be learned provided that the plastic adaptation is sufficiently slow.

Forward propagating reinforcement learning—biologically plausible learning method for multi-layer networks

We introduce a biologically plausible method of implementing reinforcement learning to multi-layer neural networks. The key idea is to spatially localize the synaptic modulation induced by reinforcement signals, proceeding downstream from the initial layer to the final layer. Since reinforcement signals are known to be broadcast signals in the actual brain, we need two key assumptions, inhibitory backward connections and bypass to output units, to spatially localize the effect of delayed reinforcement without breaking the basic laws of neurophysiology.

Adaptive learning by extremal dynamics and negative feedback

We describe a mechanism for biological learning and adaptation based on two simple principles: (i) Neuronal activity propagates only through the network's strongest synaptic connections (extremal dynamics), and (ii) the strengths of active synapses are reduced if mistakes are made, otherwise no changes occur (negative feedback). The balancing of those two tendencies typically shapes a synaptic landscape with configurations which are barely stable, and therefore highly flexible. This allows for swift adaptation to new situations. Recollection of past successes is achieved by punishing synapses which have once participated in activity associated with successful outputs much less than neurons that have never been successful. Despite its simplicity, the model can readily learn to solve complicated nonlinear tasks, even in the presence of noise. In particular, the learning time for the benchmark parity problem scales algebraically with the problem size N, with an exponent k approximately 1.4.