John H. Andreae

Detection of adaptive inverse models in the human motor system

Abstract This study aimed to find evidence for the formation of an internal inverse model of a novel visuomotor relationship for feedforward control in the brain. An experiment was carried out involving 20 normal adult subjects who performed a pursuit random tracking task with a steering wheel for input. During learning the response cursor was periodically blanked, removing all feedback about the external system (i.e., about the relationship between hand motion and response cursor motion). Results showed a transfer of learning from the unblanked runs to the blanked runs for a static nonlinear system (linear trend RMS error F (1,19)=5.05, p =0.037) thereby demonstrating adaptive feedforward control in the nervous system. No such transfer was observed for a dynamic linear system, indicating a dominant adaptive feedback control component. Results are consistent with inverse modeling and suggest a combination of feedforward and feedback adaptive control in the brain.

A learning machine with monologue

The learning machine STeLLA has been developed considerably since it was first described in 1962. Although it has not been built in entirety, it has been simulated on computers in many different forms and with many different problems; special circuits have been developed for its construction. The need to give the machine a “monologue” ability arose because the machine cannot learn to solve problems for which the input to the machine is inadequate to distinguish successive steps made by the machine. A simple example of this is the problem of learning to perform a sequence of actions when the sequence is not related to information received by the machine from its problem environment. In this paper an introductory description of the STeLLA machine is given with the help of a particular problem which is then used to illustrate the generation of control policies by a dual machine. The dual STeLLA comprises two interdependent STeLLA machines or STeLLAments; one interacts directly with the problem environment, while the second STeLLAment interacts with an auxiliary “vocal” environment to provide monologue. Monologue is used to supplement the information from the problem invironment with information from the vocal environment. The STeLLAment interacting directly with the vocal environment has its input supplemented by information from the control policy of the other machine. The two machines are co-ordinated further by giving reward to the dual machine as a whole. The procedure of counting is the kind of monologue which can be used to distinguish a sequence of steps that cannot be distinguished by successive inputs from the problem environment. This is illustrated in the paper by considering a problem that requires STeLLA “to walk across a dark courtyard”. Monologue is not restricted to such counting of steps, but can take the form of more sophisticated “symbol” sequences. By allowing an operator to inject sequences into the monologue of the machine, an elementary form of dialogue could be set up. It would be a special feature of the dialogue that no predetermined language was imposed on the participants. Examples of the use of monologue have been worked out and more complicated situations are being programmed for computer simulation both in the monologue and dialogue form.

Simulating closed- and open-loop voluntary movement: a nonlinear control-systems approach

In many recent human motor control models, including feedback-error learning and adaptive model theory (AMT), feedback control is used to correct errors while an inverse model is simultaneously tuned to provide accurate feedforward control. This popular and appealing hypothesis, based on a combination of psychophysical observations and engineering considerations, predicts that once the tuning of the inverse model is complete the role of feedback control is limited to the correction of disturbances. This hypothesis was tested by looking at the open-loop behavior of the human motor system during adaptation. An experiment was carried out involving 20 normal adult subjects who learned a novel visuomotor relationship on a pursuit tracking task with a steering wheel for input. During learning, the response cursor was periodically blanked, removing all feedback about the external system (i.e., about the relationship between hand motion and response cursor motion). Open-loop behavior was not consistent with a progressive transfer from closed to open-loop control. Our recently developed computational model of the brain-a novel nonlinear implementation of AMT-was able to reproduce the observed closed- and open-loop results. In contrast, other control-systems models exhibited only minimal feedback control following adaptation, leading to incorrect open-loop behavior. This is because our model continues to use feedback to control slow movements after adaptation is complete. This behavior enhances the internal stability of the inverse model. In summary, our computational model is currently the only motor control model able to accurately simulate the closed- and open-loop characteristics of the experimental response trajectories.

Measurement and Analysis of Single and Multiple Finger Tapping in Normal and Parkinsonian Subjects

A four-finger tapping sensor and associated software have been developed for the investigation of timing and rhythm performance and mechanisms in normal subjects and their disruption in neurological disorders. The tapping sensor comprises four electronic touch pads and pacing lights. A personal computer (PC) is used to control visual and auditory pacing, record the time and pad of each tap, and carry out several performance analyses including graphing, phase-space plots, calculation of spectra and autocorrelations, filtering and descriptive statistical analysis. A study was conducted to investigate disruptions of timing and rhythm in subjects with Parkinsons disease (PD). Seven Parkinsonian and ten control subjects undertook paced and unpaced finger tapping tests. The hastening phenomenon-tapping asynchronously at a speed faster than the pacing-was seen with similar occurrence in both normal and PD subjects and appears to be due to perceptual difficulties. No evidence was seen of an increased variability of tapping at particular frequencies, contrary to previous reports. Festinated tapping, in which subjects cycled between acceleration to near-maximum speed and abrupt slowing down, was seen only in PD subjects. As noneof these subjects showed significant hastening, it appears that hastening and festination are unrelated phenomena. Inspection of variations of finger tapping intervals gave no evidence for the presence of deterministic chaos in the control of rhythmic tapping. In speed tests performed with and without a weight attached to the finger, it was found that, for reasons which remain unclear, normal subjects increased their maximum tapping speed with the weight attached.

A New Mechanism for a Brain.

The new mechanism, PUSS, enables experience of any complex environment to be accumulated in a predictive model. PURR-PUSS is a teachable robot system based on the new mechanism. Cumulative learning is demonstrated by a detailed example. PURR-PUSS is taught to behave like a universal Turing machine.

30 to 230 megacycle pulse technique for ultrasonic absorption measurements in liquids

A proved design for a pulse technique for measuring ultrasonic absorption in liquids is described in detail. The incorporation of servo-control and special methods of data-handling have made it possible for the speed and accuracy of measurements to be increased appreciably. A collection of data obtained from pure liquids by means of the equipment is included for the benefit of other experimenters.

1.5 to 28.5 Mc/s pulse apparatus for automatic measurement of sound absorption in liquids and some results for aqueous and other solutions

The paper contains a detailed description of pulse equipment which has been built and used for measuring ultrasonic absorption in liquids. The apparatus automatically produces a paper recording from which the absorption coefficient is easily calculated. Absorption data for 17 aqueous solutions and 2 non-aqueous solutions are given. Measurements made with this apparatus and a high frequency apparatus extending to 200 Mc/s have been analysed in terms of single and double relaxation processes and the resulting parameters A, B and fc are tabulated.

THE COMPETENCE OF A MULTIPLE CONTEXT LEARNING SYSTEM

There are four ways in which a finite (and therefore realizable) robot learning system can be given beyond-finile-slutc computational power: (1) by being given “auxiliary” actions, like speech, the system achieves a beyond-finile-stale “competence”(2) by being coupled to (made “open” to) a beyond-finite-statc system, like the real world, the system attains a beyond-finite-state “performance” (3) by being limited to a finite “lifetime”, the systems finite-stale behaviour can be made indistinguishable from beyond-finite-state “competence” and “performance” and (4) by being taught the grammar of a beyond-finite-state language, the system acquires a beyond-finite-state “competence”. The “competence” and “performance” of our Multiple Context Learning System (MCLS) have been argued and demonstrated, previously, in the first three ways, even though our critics choose to ignore the record! Here, it is shown explicitly and in detail how the MCLS can learn, hold and simulate both the finite-stale controller and th...

Improving the performance of Kanerva's associate memory

A parallel associative memory first proposed by Kanerva (1988) is discussed. The major appeal of this memory is its ability to be trained very rapidly. A discrepancy between Kanervas theoretical calculation of capacity and the actual capacity is demonstrated experimentally and a corrected theory is offered. A modified method of reading from memory is suggested which results in a capacity nearly the same as that originally predicted by Kanerva. The capacity of the memory is then analyzed for a different method of writing to memory. This method increases the capacity of the memory by an order of magnitude. A further modification is suggested which increases the learning rate of this method.

Ultrasonic Relaxation in Methylene Chloride

Ultrasonic absorption measurements have been made in liquid methylene chloride between 7.5 and 210 Mc/s. These confirm recent work in the vapour by Sette, Busala and Hubbard (1955), from which it was concluded that the relaxation of the vibrational specific heat takes place in two parts. In the frequency range studied here there is a single relaxation of specific heat equal in magnitude to the vibrational specific heat less the contribution of the lowest frequency mode of vibration. The relaxation frequency is 170 Mc/s at 25°C.