Model-based closed-loop control of thalamic deep brain stimulation.

Introduction: Closed-loop control of deep brain stimulation (DBS) is beneficial for effective and automatic treatment of various neurological disorders like Parkinson's disease (PD) and essential tremor (ET). Manual (open-loop) DBS programming solely based on clinical observations relies on neurologists' expertise and patients' experience. Continuous stimulation in open-loop DBS may decrease battery life and cause side effects. On the contrary, a closed-loop DBS system uses a feedback biomarker/signal to track worsening (or improving) of patients' symptoms and offers several advantages compared to the open-loop DBS system. Existing closed-loop DBS control systems do not incorporate physiological mechanisms underlying DBS or symptoms, e.g., how DBS modulates dynamics of synaptic plasticity. Methods: In this work, we propose a computational framework for development of a model-based DBS controller where a neural model can describe the relationship between DBS and neural activity and a polynomial-based approximation can estimate the relationship between neural and behavioral activities. A controller is used in our model in a quasi-real-time manner to find DBS patterns that significantly reduce the worsening of symptoms. By using the proposed computational framework, these DBS patterns can be tested clinically by predicting the effect of DBS before delivering it to the patient. We applied this framework to the problem of finding optimal DBS frequencies for essential tremor given electromyography (EMG) recordings solely. Building on our recent network model of ventral intermediate nuclei (Vim), the main surgical target of the tremor, in response to DBS, we developed neural model simulation in which physiological mechanisms underlying Vim-DBS are linked to symptomatic changes in EMG signals. By using a proportional-integral-derivative (PID) controller, we showed that a closed-loop system can track EMG signals and adjust the stimulation frequency of Vim-DBS so that the power of EMG reaches a desired control target. Results and discussion: We demonstrated that the model-based DBS frequency aligns well with that used in clinical studies. Our model-based closed-loop system is adaptable to different control targets and can potentially be used for different diseases and personalized systems.

Adaptive unscented Kalman filter for neuronal state and parameter estimation.

Data assimilation techniques for state and parameter estimation are frequently applied in the context of computational neuroscience. In this work, we show how an adaptive variant of the unscented Kalman filter (UKF) performs on the tracking of a conductance-based neuron model. Unlike standard recursive filter implementations, the robust adaptive unscented Kalman filter (RAUKF) jointly estimates the states and parameters of the neuronal model while adjusting noise covariance matrices online based on innovation and residual information. We benchmark the adaptive filter's performance against existing nonlinear Kalman filters and explore the sensitivity of the filter parameters to the system being modelled. To evaluate the robustness of the proposed solution, we simulate practical settings that challenge tracking performance, such as a model mismatch and measurement faults. Compared to standard variants of the Kalman filter the adaptive variant implemented here is more accurate and robust to faults.

At-scale lift experiments modeling dragonfly forewings.

An experimental investigation of the lift performance of an artificial platform at the scale of the dragonfly species Sympetrum sanguineum is presented. The platform, as well as the lift sensor, was custom designed and built. The flapping mechanism consisted of a piezoelectric bending-beam actuator, a transmission using carbon-fiber elements and polymide-film joints, and wings constructed of polyester film with a carbon-fiber support structure. The flapping kinematics of the Sympetrum sanguineum was replicated as closely as possible although only a pair of forewings were used in these experiments. The lift generated, when accounting for the addition of a pair of hindwings, is predicted to be sufficient to allow for the hovering of a dragonfly. The results, the first at-scale fully transient measurements of artificial dragonfly forewings, show that the lift curves quantitatively as well as qualitatively validate existing two-dimensional and three-dimensional computer simulations of dragonfly forewings.

On The Evolutionary Origin of Symbolic Communication.

The emergence of symbolic communication is often cited as a critical step in the evolution of Homo sapiens, language, and human-level cognition. It is a widely held assumption that humans are the only species that possess natural symbolic communication schemes, although a variety of other species can be taught to use symbols. The origin of symbolic communication remains a controversial open problem, obfuscated by the lack of a fossil record. Here we demonstrate an unbroken evolutionary pathway from a population of initially noncommunicating robots to the spontaneous emergence of symbolic communication. Robots evolve in a simulated world and are supplied with only a single channel of communication. When their ability to reproduce is motivated by the need to find a mate, robots evolve indexical communication schemes from initially noncommunicating populations in 99% of all experiments. Furthermore, 9% of the populations evolve a symbolic communication scheme allowing pairs of robots to exchange information about two independent spatial dimensions over a one-dimensional channel, thereby increasing their chance of reproduction. These results suggest that the ability for symbolic communication could have emerged spontaneously under natural selection, without requiring cognitive preadaptations or preexisting iconic communication schemes as previously conjectured.

Synthesis of recurrent neural networks for dynamical system simulation.

We review several of the most widely used techniques for training recurrent neural networks to approximate dynamical systems, then describe a novel algorithm for this task. The algorithm is based on an earlier theoretical result that guarantees the quality of the network approximation. We show that a feedforward neural network can be trained on the vector-field representation of a given dynamical system using backpropagation, then recast it as a recurrent network that replicates the original system's dynamics. After detailing this algorithm and its relation to earlier approaches, we present numerical examples that demonstrate its capabilities. One of the distinguishing features of our approach is that both the original dynamical systems and the recurrent networks that simulate them operate in continuous time.

Tackling learning intractability through topological organization and regulation of cortical networks.

A key challenge in evolving control systems for robots using neural networks is training tractability. Evolving monolithic fixed topology neural networks is shown to be intractable with limited supervision in high dimensional search spaces. Common strategies to overcome this limitation are to provide more supervision by encouraging particular solution strategies, manually decomposing the task and segmenting the search space and network. These strategies require a supervisor with domain knowledge and may not be feasible for difficult tasks where novel concepts are required. The alternate strategy is to use self-organized task decomposition to solve difficult tasks with limited supervision. The artificial neural tissue (ANT) approach presented here uses self-organized task decomposition to solve tasks. ANT inspired by neurobiology combines standard neural networks with a novel wireless signaling scheme modeling chemical diffusion of neurotransmitters. These chemicals are used to dynamically activate and inhibit wired network of neurons using a coarse-coding framework. Using only a global fitness function that does not encourage a predefined solution, modular networks of neurons are shown to self-organize and perform task decomposition. This approach solves the sign-following task found to be intractable with conventional fixed and variable topology networks. In this paper, key attributes of the ANT architecture that perform self-organized task decomposition are shown. The architecture is robust and scalable to number of neurons, synaptic connections, and initialization parameters.