Showing papers by "Byron M. Yu published in 2005"

Extracting Dynamical Structure Embedded in Neural Activity

Abstract: Spiking activity from neurophysiological experiments often exhibits dynamics beyond that driven by external stimulation, presumably reflecting the extensive recurrence of neural circuitry. Characterizing these dynamics may reveal important features of neural computation, particularly during internally-driven cognitive operations. For example, the activity of premotor cortex (PMd) neurons during an instructed delay period separating movement-target specification and a movement-initiation cue is believed to be involved in motor planning. We show that the dynamics underlying this activity can be captured by a low-dimensional non-linear dynamical systems model, with underlying recurrent structure and stochastic point-process output. We present and validate latent variable methods that simultaneously estimate the system parameters and the trial-by-trial dynamical trajectories. These methods are applied to characterize the dynamics in PMd data recorded from a chronically-implanted 96-electrode array while monkeys perform delayed-reach tasks.

Extracting dynamical structure embedded in neural activity

Abstract: Spiking activity from neurophysiological experiments often exhibits dynamics beyond that driven by external stimulation, presumably reflecting the extensive recurrence of neural circuitry. Characterizing these dynamics may reveal important features of neural computation, particularly during internally-driven cognitive operations. For example, the activity of premotor cortex (PMd) neurons during an instructed delay period separating movement-target specification and a movement-initiation cue is believed to be involved in motor planning. We show that the dynamics underlying this activity can be captured by a low-dimensional non-linear dynamical systems model, with underlying recurrent structure and stochastic point-process output. We present and validate latent variable methods that simultaneously estimate the system parameters and the trial-by-trial dynamical trajectories. These methods are applied to characterize the dynamics in PMd data recorded from a chronically-implanted 96-electrode array while monkeys perform delayed-reach tasks.

A High Performance Neurally-Controlled Cursor Positioning System

Abstract: Prior work has shown that neural activity from the primate brain can maneuver a computer cursor to specified visual targets. This cursor movement can take over a second, longer than the time for an arm reach to the same location. We asked if this acquisition time could be reduced, thereby increasing the number of targets that could be hit per second. We implemented a system that positions a prosthetic cursor at discrete locations, based on pre-movement neural activity in rhesus monkeys. Using a delayed center-out reaching task with several different target layouts, neural activity was simultaneously recorded from an electrode array implanted in the dorsal pre-motor cortex. We designed a target prediction algorithm based on maximum-likelihood models (using Gaussian or Poisson distributions) to decode the upcoming reach target in real-time. During cursor trials, the algorithm predicted the most likely reach target using 50-275 ms of delay activity starting at least 150 ms after target onset If the target prediction was correct, a cursor was positioned and the monkey received a reward. The performance of the system was evaluated based on the accuracy of decoded targets and speed at which targets were decoded, both of which were consolidated with an information theoretic analysis. The maximum average sustained rate of target acquisition was 43 targets per second obtained with a 2 target layout and 50 ms of delay activity. The maximum information transfer rate calculated for the system was 6.5 bps obtained with an 8 target layout and 100 ms of delay activity

Mixture of trajectory models for neural decoding of goal-directed movements

Abstract: Probabilistic decoding techniques have been used successfully to infer time-evolving physical state, such as arm trajectory or the path of a foraging rat, from neural data. A vital element of such decoders is the trajectory model, expressing knowledge about the statistical regularities of the movements. Unfortunately, trajectory models that both 1) accurately describe the movement statistics and 2) admit decoders with relatively low computational demands can be hard to construct. Simple models are computationally inexpensive, but often inaccurate. More complex models may gain accuracy, but at the expense of higher computational cost, hindering their use for real-time decoding. Here, we present a new general approach to defining trajectory models that simultaneously meets both requirements. The core idea is to combine simple trajectory models, each accurate within a limited regime of movement, in a probabilistic mixture of trajectory models (MTM). We demonstrate the utility of the approach by using an MTM decoder to infer goal-directed reaching movements to multiple discrete goals from multi-electrode neural data recorded in monkey motor and premotor cortex. Compared with decoders using simpler trajectory models, the MTM decoder reduced the decoding error by 38 (48) percent in two monkeys using 98 (99) units, without a necessary increase in running time. When available, prior information about the identity of the upcoming reach goal can be incorporated in a principled way, further reducing the decoding error by 20 (11) percent. Taken together, these advances should allow prosthetic cursors or limbs to be moved more accurately toward intended reach goals.

Trial - by - trial mean normalization improves plan period reach target decoding

Abstract: leftward reaches have a mean firing rate of 3.4 1.1). To explore this finding, two different multivariate Poisson models were generated. The first model assumes independence between the firing rate of each unit. The second model assumes the firing rate of each unit is coupled to the population's mean firing rate; accordingly, the model scales the expected firing rate of each unit by this mean firing rate on each trial. These models were fit offline to the first 200 trials of the day and were tested on trials from the remainder of the day. For monkey G (H), when decoding on 150ms of delay period activity, the first and second model correctly classify reach direction on 55% (75%) and 66% (78%) of the trials, respectively (chance=12.5%). Across multiple daily data sets collected over a period of several weeks, the second model consistently outperforms the first model. If we adjust the expected firing rate of individual neurons according to mean firing rate, we may improve discrete classification and continuous estimation of reaches from neural activity. Improving these processes will move us closer to realizing effective neural prosthetic systems.

Intracortical communication prosthesis design

Abstract: Cortical neural activity can be used to position computer cursors on visually presented targets. We wished to investigate the following questions central to the clinical viability of such prostheses: how quickly and accurately can targets be selected; how performance differs when running online; and whether performance is limited by electrode count or other factors. Experiments were performed with two rhesus macaques trained on an instructed-delay center-out reach task (2, 4, 8 or 16 target locations). Single and multi-unit activity was recorded from a 96-electrode array in premotor cortex. The desired target was estimated using 50-275 ms of delay activity (Tint epoch) starting 150 ms after target onset (Tskip epoch). If the target prediction was correct a cursor was positioned, the monkey received a reward, and the next target appeared immediately. Tskip was estimated in control experiments in order to exclude noise or visual transients from the Tint estimation window. With one monkey, we recently performed a series of online experiments where Tint was swept to understand the effect of this parameter on the maximum information transmission rate (bits/s) between the requested and the estimated target. Offline analyses using data from trials with targets presented ~2 sec apart indicated that 8.1 bits/s could be achieved under these conditions. For online experiments, the maximum average sustained performance achieved was 6.5 bits/s (8 targets, Tint=100 ms). The difference between offline (8.1) and online (6.5) performance appears to result from changes in neural tuning curves when targets are presented in rapid succession (Kalmar et al. in this volume). We also report results from modeling efforts to understand the prospect for higher information throughput afforded by adding electrodes to the system. This suggests that the maximum achievable throughput (~10 bits/s) is limited by task and decoding design, as opposed to the total number of recording electrodes.

Extracting Dynamical Structure Embedded in Motor Preparatory Activity

Abstract: At present, the best view of the activity of a neural circuit is provided by multiple-electrode extracellular recording technologies. These technologies are able to simultaneously measure spike trains from up to a few hundred cells in one or more brain areas during each trial. However, characterizing the dynamics of the circuit from such data is made difficult, both by the point-like nature of the spikes and by the variability in the responses to identical repeated trials. By their nature, spikes give us only an occasional view of the process from which they are generated.